diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_0.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_0.log new file mode 100644 index 0000000000000000000000000000000000000000..5dd13a0ba8694a7421416594bc4ffd4a25353a52 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_0.log @@ -0,0 +1,7422 @@ +2026-05-29 20:46:14.505 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-29 20:46:14.505 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-29 20:46:16.602 | INFO | __main__:print_training_configuration:963 - SYSTEM & HARDWARE CONFIGURATION +2026-05-29 20:46:16.604 | INFO | __main__:print_training_configuration:964 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.604 | INFO | __main__:print_training_configuration:965 - Number of GPUs : 64 +2026-05-29 20:46:16.605 | INFO | __main__:print_training_configuration:966 - Local rank : 0 +2026-05-29 20:46:16.605 | INFO | __main__:print_training_configuration:967 - Global rank : 0 +2026-05-29 20:46:16.606 | INFO | __main__:print_training_configuration:968 - Device : cuda:0 +2026-05-29 20:46:16.606 | INFO | __main__:print_training_configuration:969 - Master weight dtype : torch.float32 +2026-05-29 20:46:16.606 | INFO | __main__:print_training_configuration:971 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.607 | INFO | __main__:print_training_configuration:972 - MODEL PARAMETERS & ARCHITECTURE +2026-05-29 20:46:16.607 | INFO | __main__:print_training_configuration:973 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.618 | INFO | __main__:print_training_configuration:976 - Number of total : 8,326,608,160 +2026-05-29 20:46:16.619 | INFO | __main__:print_training_configuration:977 - Total trainable parameters : 8,326,608,160 +2026-05-29 20:46:16.619 | INFO | __main__:print_training_configuration:978 - DIT model root : ./ckpts +2026-05-29 20:46:16.619 | INFO | __main__:print_training_configuration:979 - Transformer version : 480p_t2v +2026-05-29 20:46:16.620 | INFO | __main__:print_training_configuration:980 - Attention mode : flash +2026-05-29 20:46:16.620 | INFO | __main__:print_training_configuration:981 - Gradient checkpointing : True +2026-05-29 20:46:16.621 | INFO | __main__:print_training_configuration:982 - FSDP enabled : True +2026-05-29 20:46:16.621 | INFO | __main__:print_training_configuration:983 - LoRA enabled : False +2026-05-29 20:46:16.621 | INFO | __main__:print_training_configuration:984 - Glyph byT5 v2 : True +2026-05-29 20:46:16.622 | INFO | __main__:print_training_configuration:985 - ByT5 max length : 256 +2026-05-29 20:46:16.622 | INFO | __main__:print_training_configuration:987 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.623 | INFO | __main__:print_training_configuration:988 - DISTRIBUTED TRAINING CONFIGURATION +2026-05-29 20:46:16.623 | INFO | __main__:print_training_configuration:989 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.623 | INFO | __main__:print_training_configuration:990 - World size : 64 +2026-05-29 20:46:16.624 | INFO | __main__:print_training_configuration:991 - DP degree : 32 +2026-05-29 20:46:16.624 | INFO | __main__:print_training_configuration:992 - DP rank : 0 +2026-05-29 20:46:16.625 | INFO | __main__:print_training_configuration:993 - SP size : 2 +2026-05-29 20:46:16.625 | INFO | __main__:print_training_configuration:994 - SP rank : 0 +2026-05-29 20:46:16.625 | INFO | __main__:print_training_configuration:996 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.626 | INFO | __main__:print_training_configuration:997 - BATCH SIZE & DATA FLOW CONFIGURATION +2026-05-29 20:46:16.626 | INFO | __main__:print_training_configuration:998 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.627 | INFO | __main__:print_training_configuration:999 - Per GPU batch size : 2 +2026-05-29 20:46:16.627 | INFO | __main__:print_training_configuration:1000 - DP degree batch size : 64 +2026-05-29 20:46:16.627 | INFO | __main__:print_training_configuration:1001 - Gradient accumulation steps : 1 +2026-05-29 20:46:16.628 | INFO | __main__:print_training_configuration:1002 - Num generations : 4 +2026-05-29 20:46:16.628 | INFO | __main__:print_training_configuration:1003 - Use same noise : False +2026-05-29 20:46:16.629 | INFO | __main__:print_training_configuration:1004 - Num workers : 4 +2026-05-29 20:46:16.629 | INFO | __main__:print_training_configuration:1005 - Prefetch factor : 2 +2026-05-29 20:46:16.629 | INFO | __main__:print_training_configuration:1007 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.630 | INFO | __main__:print_training_configuration:1008 - OPTIMIZATION CONFIGURATION +2026-05-29 20:46:16.630 | INFO | __main__:print_training_configuration:1009 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.631 | INFO | __main__:print_training_configuration:1010 - Optimizer : muon +2026-05-29 20:46:16.631 | INFO | __main__:print_training_configuration:1011 - Learning rate : 1e-05 +2026-05-29 20:46:16.631 | INFO | __main__:print_training_configuration:1012 - Weight decay : 0.01 +2026-05-29 20:46:16.632 | INFO | __main__:print_training_configuration:1013 - Max gradient norm : 1.0 +2026-05-29 20:46:16.632 | INFO | __main__:print_training_configuration:1014 - LR scheduler : constant +2026-05-29 20:46:16.632 | INFO | __main__:print_training_configuration:1015 - Warmup steps : 500 +2026-05-29 20:46:16.633 | INFO | __main__:print_training_configuration:1016 - Max training steps : 10000 +2026-05-29 20:46:16.633 | INFO | __main__:print_training_configuration:1018 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.634 | INFO | __main__:print_training_configuration:1019 - INFERENCE & VALIDATION CONFIGURATION +2026-05-29 20:46:16.634 | INFO | __main__:print_training_configuration:1020 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1021 - Validation interval : 0 +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1022 - Validate at step 0 : False +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1023 - Validation video length : 121 +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1024 - Validation global seed : 930 +2026-05-29 20:46:16.636 | INFO | __main__:print_training_configuration:1025 - Validation inference steps : 40 +2026-05-29 20:46:16.636 | INFO | __main__:print_training_configuration:1026 - Validation aspect ratio : 16:9 +2026-05-29 20:46:16.637 | INFO | __main__:print_training_configuration:1027 - Validation fixed size : 480x864 +2026-05-29 20:46:16.637 | INFO | __main__:print_training_configuration:1028 - Validation prompt rewrite : False +2026-05-29 20:46:16.638 | INFO | __main__:print_training_configuration:1029 - Validation flow shift : 5.0 +2026-05-29 20:46:16.638 | INFO | __main__:print_training_configuration:1030 - Sampling CFG scale : 6.0 +2026-05-29 20:46:16.638 | INFO | __main__:print_training_configuration:1031 - Sampling eta : 0.5 +2026-05-29 20:46:16.639 | INFO | __main__:print_training_configuration:1032 - Sampling steps (GRPO rollout) : 20 +2026-05-29 20:46:16.639 | INFO | __main__:print_training_configuration:1033 - Token credit loss mode : video_adv +2026-05-29 20:46:16.640 | INFO | __main__:print_training_configuration:1034 - Token credit ratio mode : uniform_weighted_logprob +2026-05-29 20:46:16.640 | INFO | __main__:print_training_configuration:1035 - Token credit shuffle weights : False +2026-05-29 20:46:16.640 | INFO | __main__:print_training_configuration:1036 - Token credit source model : None +2026-05-29 20:46:16.641 | INFO | __main__:print_training_configuration:1037 - Token credit advantage source : token +2026-05-29 20:46:16.641 | INFO | __main__:print_training_configuration:1038 - Random train timestep count : False +2026-05-29 20:46:16.641 | INFO | __main__:print_training_configuration:1039 - Random train timestep range : 1-20 +2026-05-29 20:46:16.642 | INFO | __main__:print_training_configuration:1043 - SDE type : sage_grpo +2026-05-29 20:46:16.642 | INFO | __main__:print_training_configuration:1045 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.643 | INFO | __main__:print_training_configuration:1046 - GRPO & KL CONFIGURATION +2026-05-29 20:46:16.643 | INFO | __main__:print_training_configuration:1047 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.643 | INFO | __main__:print_training_configuration:1048 - Reward model : video_score2 +2026-05-29 20:46:16.644 | INFO | __main__:print_training_configuration:1049 - Reward gradient credit : True +2026-05-29 20:46:16.644 | INFO | __main__:print_training_configuration:1050 - Reward credit frames : 8 +2026-05-29 20:46:16.645 | INFO | __main__:print_training_configuration:1051 - Reward credit max questions : 1 +2026-05-29 20:46:16.645 | INFO | __main__:print_training_configuration:1052 - Reward credit metric : weighted +2026-05-29 20:46:16.645 | INFO | __main__:print_training_configuration:1063 - Reward checkpoint mode : v3 +2026-05-29 20:46:16.646 | INFO | __main__:print_training_configuration:1064 - KL weight : 1e-05 +2026-05-29 20:46:16.646 | INFO | __main__:print_training_configuration:1065 - KL coef : 1e-07 +2026-05-29 20:46:16.647 | INFO | __main__:print_training_configuration:1066 - Use moving KL : True +2026-05-29 20:46:16.647 | INFO | __main__:print_training_configuration:1067 - Update ref model step : 10 +2026-05-29 20:46:16.647 | INFO | __main__:print_training_configuration:1068 - Use dual KL : True +2026-05-29 20:46:16.648 | INFO | __main__:print_training_configuration:1069 - Dual KL moving weight : 1.0 +2026-05-29 20:46:16.648 | INFO | __main__:print_training_configuration:1070 - Dual KL step weight : 0.1 +2026-05-29 20:46:16.648 | INFO | __main__:print_training_configuration:1071 - KL compute mode : rollout_phase +2026-05-29 20:46:16.649 | INFO | __main__:print_training_configuration:1072 - Use grad balancing : True +2026-05-29 20:46:16.649 | INFO | __main__:print_training_configuration:1073 - Enable timestep permutation : True +2026-05-29 20:46:16.650 | INFO | __main__:print_training_configuration:1074 - Reference mode offload : True +2026-05-29 20:46:16.650 | INFO | __main__:print_training_configuration:1076 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.651 | INFO | __main__:print_training_configuration:1077 - LOGGING & CHECKPOINTING CONFIGURATION +2026-05-29 20:46:16.651 | INFO | __main__:print_training_configuration:1078 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.651 | INFO | __main__:print_training_configuration:1079 - Output directory : ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-29 20:46:16.652 | INFO | __main__:print_training_configuration:1080 - Checkpoints directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-29 20:46:16.652 | INFO | __main__:print_training_configuration:1081 - Checkpoint interval : 100 +2026-05-29 20:46:16.653 | INFO | __main__:print_training_configuration:1082 - Timer checkpoint interval : 25 +2026-05-29 20:46:16.653 | INFO | __main__:print_training_configuration:1083 - Auto resume from timer : True +2026-05-29 20:46:16.653 | INFO | __main__:print_training_configuration:1084 - Timer checkpoint directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer +2026-05-29 20:46:16.654 | INFO | __main__:print_training_configuration:1085 - Log interval : 1 +2026-05-29 20:46:16.654 | INFO | __main__:print_training_configuration:1086 - Resume checkpoint : None +2026-05-29 20:46:16.655 | INFO | __main__:print_training_configuration:1087 - Train CSV : ['/nfs/ywang29/Reward_finetuning/SAGE-GRPO/assets/demo_train_decomposed_binary5.csv'] +2026-05-29 20:46:16.655 | INFO | __main__:print_training_configuration:1088 - Valid CSV : assets/demo_sample.csv +2026-05-29 20:46:16.655 | INFO | __main__:print_training_configuration:1089 - Seed : 42 +2026-05-29 20:46:16.656 | INFO | __main__:print_training_configuration:1090 - WandB enabled : True +2026-05-29 20:46:16.656 | INFO | __main__:print_training_configuration:1092 - WandB project : SAGE-GRPO +2026-05-29 20:46:16.657 | INFO | __main__:print_training_configuration:1093 - WandB run name : yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-29 20:46:16.657 | INFO | __main__:__init__:951 - [Effective Config] world_size=64, rank=0, local_rank=0, sp_size=2, dp_size=32, dp_rank=0, sp_rank=0, batch_size=2, num_generations=4, grad_accum=1 +2026-05-29 20:46:16.657 | INFO | __main__:_build_video_dataset:2038 - [Effective GRPO Grouping] video_batch_size=2, dp_replicas=32, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, product=64 +2026-05-29 20:46:17.315 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:46:17.318 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:46:17.319 | INFO | __main__:_train_grpo:2121 - Starting GRPO training... +2026-05-29 20:46:17.319 | INFO | __main__:_train_grpo:2122 - Max steps: 10000 +2026-05-29 20:46:17.320 | INFO | __main__:_train_grpo:2123 - Batch size: 2 +2026-05-29 20:46:17.320 | INFO | __main__:_train_grpo:2124 - Learning rate: 1e-05 +2026-05-29 20:46:17.512 | INFO | __main__:_resolve_resume_checkpoint:1820 - No timer checkpoint found at s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer; starting from scratch. +2026-05-29 20:46:19.144 | INFO | __main__:_train_grpo:2151 - [FLAG] post_train train path -> train_one_step +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 20:51:46.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28576_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 20:53:30.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:30.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:38.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28577_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 20:55:07.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 20:55:17.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 0, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.122 | INFO | __main__:_log_precision_diagnostics_once:1624 - [PrecisionDiag] trainable param dtypes: ['torch.float32'] +2026-05-29 21:06:57.127 | INFO | __main__:_log_precision_diagnostics_once:1627 - [PrecisionDiag] Using composable FSDP2 mixed precision (param_dtype=bf16, reduce_dtype=float32). This path does not maintain an explicit FP32 master-weight copy for parameters. +2026-05-29 21:06:57.130 | INFO | __main__:_log_precision_diagnostics_once:1640 - [PrecisionDiag] optimizer tensor-state dtypes: ['torch.float32'] (count=2804) +2026-05-29 21:06:57.130 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 1 train_steps 40 {'step': '1/10000', 'learning_rate': '0.00001000', 'loss': '2.561e-08', 'grad_norm': '0.00341101', 'KL_weight': '0.00000396', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '337.42507935', 'grpo_loop_time': '453.07379150', 'ref_model_time': '124.71535492', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87055492', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.09570312', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.526e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.164e-09', 'kl_loss': '0.00702148', 'kl_moving': '0.00702148', 'kl_stepwise': '-1.971e-08', 'gathered_avg_reward_mean': '9.09570312', 'gathered_avg_reward_std': '0.92303908', 'gathered_video_score2_mq_reward_mean': '3.12341309', 'gathered_video_score2_mq_reward_std': '0.36417580', 'gathered_video_score2_overall_reward_mean': '9.09570312', 'gathered_video_score2_overall_reward_std': '0.91749281', 'gathered_video_score2_ta_reward_mean': '3.20593262', 'gathered_video_score2_ta_reward_std': '0.45509985', 'gathered_video_score2_vq_reward_mean': '2.76635742', 'gathered_video_score2_vq_reward_std': '0.42974457', 'gathered_ori_avg_reward_mean': '9.09570312', 'gathered_ori_avg_reward_std': '0.92303908'} +2026-05-29 21:06:57.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 21:06:57.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:04.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_0/index_6771_seed_33047_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 21:08:19.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:19.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:09:39.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_0/index_6771_seed_33048_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:10:01.049 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 21:10:01.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.442 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.118 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 2 train_steps 80 {'step': '2/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-08', 'grad_norm': '0.00527330', 'KL_weight': '0.00000792', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.55889893', 'grpo_loop_time': '429.29833984', 'ref_model_time': '121.69975281', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88020623', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.49121094', 'token_credit_video_advantage_mean': '-1.304e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.469e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00745757', 'kl_moving': '0.00745757', 'kl_stepwise': '-1.912e-08', 'gathered_avg_reward_mean': '8.48840332', 'gathered_avg_reward_std': '0.97872603', 'gathered_video_score2_mq_reward_mean': '2.98486328', 'gathered_video_score2_mq_reward_std': '0.24914357', 'gathered_video_score2_overall_reward_mean': '8.49121094', 'gathered_video_score2_overall_reward_std': '0.98025036', 'gathered_video_score2_ta_reward_mean': '3.03271484', 'gathered_video_score2_ta_reward_std': '0.56645823', 'gathered_video_score2_vq_reward_mean': '2.47082520', 'gathered_video_score2_vq_reward_std': '0.51014757', 'gathered_ori_avg_reward_mean': '8.48840332', 'gathered_ori_avg_reward_std': '0.97872615'} +2026-05-29 21:21:17.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 21:21:17.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:25.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_0/index_5387_seed_9522_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 21:22:37.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:37.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:23:53.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_0/index_5387_seed_9523_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 21:24:14.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:24:15.086 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 21:24:15.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.183 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 3 train_steps 120 {'step': '3/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-08', 'grad_norm': '0.00594226', 'KL_weight': '0.00000120', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.54541016', 'grpo_loop_time': '431.75509644', 'ref_model_time': '122.63415527', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86326522', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.87109375', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.461e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00789424', 'kl_moving': '0.00789424', 'kl_stepwise': '-1.944e-08', 'gathered_avg_reward_mean': '8.87011719', 'gathered_avg_reward_std': '1.43640399', 'gathered_video_score2_mq_reward_mean': '3.07714844', 'gathered_video_score2_mq_reward_std': '0.39823520', 'gathered_video_score2_overall_reward_mean': '8.87109375', 'gathered_video_score2_overall_reward_std': '1.43691075', 'gathered_video_score2_ta_reward_mean': '3.22021484', 'gathered_video_score2_ta_reward_std': '0.57646453', 'gathered_video_score2_vq_reward_mean': '2.57275391', 'gathered_video_score2_vq_reward_std': '0.61644828', 'gathered_ori_avg_reward_mean': '8.87011719', 'gathered_ori_avg_reward_std': '1.43640399'} +2026-05-29 21:35:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 21:35:35.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_0/index_3341_seed_10098_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 21:36:59.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:59.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:14.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_0/index_3341_seed_10099_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:38:35.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 21:38:36.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:55.945 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 4 train_steps 160 {'step': '4/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00481455', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.85662842', 'grpo_loop_time': '433.13183594', 'ref_model_time': '120.63969421', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85966229', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.63427734', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.411e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-09', 'kl_loss': '0.00768832', 'kl_moving': '0.00768833', 'kl_stepwise': '-1.903e-08', 'gathered_avg_reward_mean': '8.63732910', 'gathered_avg_reward_std': '1.00043941', 'gathered_video_score2_mq_reward_mean': '3.07592773', 'gathered_video_score2_mq_reward_std': '0.24953979', 'gathered_video_score2_overall_reward_mean': '8.63427734', 'gathered_video_score2_overall_reward_std': '1.00415218', 'gathered_video_score2_ta_reward_mean': '2.90490723', 'gathered_video_score2_ta_reward_std': '0.65815181', 'gathered_video_score2_vq_reward_mean': '2.65649414', 'gathered_video_score2_vq_reward_std': '0.50694215', 'gathered_ori_avg_reward_mean': '8.63732910', 'gathered_ori_avg_reward_std': '1.00043941'} +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 21:49:56.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_0/index_4150_seed_30223_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 21:51:20.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:20.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:52:35.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_0/index_4150_seed_30224_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:53:00.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 21:53:01.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:17.719 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 5 train_steps 200 {'step': '5/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-10', 'grad_norm': '0.00485901', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '310.16647339', 'grpo_loop_time': '430.86810303', 'ref_model_time': '123.64433289', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85819942', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.74902344', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-10', 'kl_loss': '0.00733614', 'kl_moving': '0.00733614', 'kl_stepwise': '-1.927e-08', 'gathered_avg_reward_mean': '8.75109863', 'gathered_avg_reward_std': '0.95966816', 'gathered_video_score2_mq_reward_mean': '3.05737305', 'gathered_video_score2_mq_reward_std': '0.25609118', 'gathered_video_score2_overall_reward_mean': '8.74902344', 'gathered_video_score2_overall_reward_std': '0.95777184', 'gathered_video_score2_ta_reward_mean': '3.02331543', 'gathered_video_score2_ta_reward_std': '0.50272673', 'gathered_video_score2_vq_reward_mean': '2.67041016', 'gathered_video_score2_vq_reward_std': '0.48110524', 'gathered_ori_avg_reward_mean': '8.75109863', 'gathered_ori_avg_reward_std': '0.95966822'} +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 22:04:18.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:26.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_0/index_7254_seed_11671_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 22:05:40.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:40.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:06:52.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_0/index_7254_seed_11672_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:07:14.408 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 22:07:15.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.080 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 6 train_steps 240 {'step': '6/10000', 'learning_rate': '0.00001000', 'loss': '-1.024e-08', 'grad_norm': '0.00386598', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '300.63519287', 'grpo_loop_time': '432.99191284', 'ref_model_time': '122.07182312', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85973650', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20605469', 'token_credit_video_advantage_mean': '1.164e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.456e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.00697387', 'kl_moving': '0.00697387', 'kl_stepwise': '-1.984e-08', 'gathered_avg_reward_mean': '9.20678711', 'gathered_avg_reward_std': '1.04540813', 'gathered_video_score2_mq_reward_mean': '3.17199707', 'gathered_video_score2_mq_reward_std': '0.37214717', 'gathered_video_score2_overall_reward_mean': '9.20605469', 'gathered_video_score2_overall_reward_std': '1.04735184', 'gathered_video_score2_ta_reward_mean': '3.16223145', 'gathered_video_score2_ta_reward_std': '0.49404749', 'gathered_video_score2_vq_reward_mean': '2.87255859', 'gathered_video_score2_vq_reward_std': '0.53518456', 'gathered_ori_avg_reward_mean': '9.20678711', 'gathered_ori_avg_reward_std': '1.04540825'} +2026-05-29 22:18:35.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 22:18:35.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:42.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_0/index_3113_seed_32440_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 22:19:57.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:57.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:14.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_0/index_3113_seed_32441_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:21:37.409 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 22:21:38.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.564 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 7 train_steps 280 {'step': '7/10000', 'learning_rate': '0.00001000', 'loss': '-1.304e-08', 'grad_norm': '0.00388570', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '308.77267456', 'grpo_loop_time': '431.44375610', 'ref_model_time': '124.55293274', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83565873', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.35742188', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.486e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.304e-08', 'kl_loss': '0.00744503', 'kl_moving': '0.00744503', 'kl_stepwise': '-1.959e-08', 'gathered_avg_reward_mean': '8.35925293', 'gathered_avg_reward_std': '1.50343728', 'gathered_video_score2_mq_reward_mean': '2.96813965', 'gathered_video_score2_mq_reward_std': '0.46372697', 'gathered_video_score2_overall_reward_mean': '8.35742188', 'gathered_video_score2_overall_reward_std': '1.50408256', 'gathered_video_score2_ta_reward_mean': '2.83105469', 'gathered_video_score2_ta_reward_std': '0.88636428', 'gathered_video_score2_vq_reward_mean': '2.56005859', 'gathered_video_score2_vq_reward_std': '0.51200545', 'gathered_ori_avg_reward_mean': '8.35925293', 'gathered_ori_avg_reward_std': '1.50343728'} +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 22:32:58.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:58.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:06.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_0/index_7447_seed_24478_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 22:34:24.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:24.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:35.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_0/index_7447_seed_24479_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:36:00.106 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 22:36:00.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:19.989 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 8 train_steps 320 {'step': '8/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00616889', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '306.71383667', 'grpo_loop_time': '431.33209229', 'ref_model_time': '123.09277344', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88111198', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.92529297', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.451e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00768356', 'kl_moving': '0.00768356', 'kl_stepwise': '-1.921e-08', 'gathered_avg_reward_mean': '8.92480469', 'gathered_avg_reward_std': '1.02804947', 'gathered_video_score2_mq_reward_mean': '3.10607910', 'gathered_video_score2_mq_reward_std': '0.32252240', 'gathered_video_score2_overall_reward_mean': '8.92529297', 'gathered_video_score2_overall_reward_std': '1.02807295', 'gathered_video_score2_ta_reward_mean': '3.10229492', 'gathered_video_score2_ta_reward_std': '0.44826606', 'gathered_video_score2_vq_reward_mean': '2.71643066', 'gathered_video_score2_vq_reward_std': '0.52468896', 'gathered_ori_avg_reward_mean': '8.92480469', 'gathered_ori_avg_reward_std': '1.02804947'} +2026-05-29 22:47:20.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 22:47:20.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_0/index_9339_seed_18934_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 22:48:45.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:45.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:49:56.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_0/index_9339_seed_18935_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:50:19.472 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 22:50:20.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.626 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 9 train_steps 360 {'step': '9/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00408050', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.67916870', 'grpo_loop_time': '433.28607178', 'ref_model_time': '126.19694519', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86826396', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.96582031', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.00682848', 'kl_moving': '0.00682848', 'kl_stepwise': '-1.871e-08', 'gathered_avg_reward_mean': '8.96484375', 'gathered_avg_reward_std': '0.92013127', 'gathered_video_score2_mq_reward_mean': '3.19897461', 'gathered_video_score2_mq_reward_std': '0.34432927', 'gathered_video_score2_overall_reward_mean': '8.96582031', 'gathered_video_score2_overall_reward_std': '0.91617543', 'gathered_video_score2_ta_reward_mean': '3.04077148', 'gathered_video_score2_ta_reward_std': '0.55036461', 'gathered_video_score2_vq_reward_mean': '2.72509766', 'gathered_video_score2_vq_reward_std': '0.44666794', 'gathered_ori_avg_reward_mean': '8.96484375', 'gathered_ori_avg_reward_std': '0.92013127'} +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 23:01:44.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_0/index_3943_seed_33139_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 23:03:12.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:03:12.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:23.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_0/index_3943_seed_33140_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:04:48.291 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 23:04:49.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:05.282 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 10 train_steps 400 {'step': '10/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00475273', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '310.86996460', 'grpo_loop_time': '430.13256836', 'ref_model_time': '124.42875671', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86619067', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.75146484', 'token_credit_video_advantage_mean': '-1.304e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.414e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00763092', 'kl_moving': '0.00763092', 'kl_stepwise': '-1.971e-08', 'gathered_avg_reward_mean': '8.75024414', 'gathered_avg_reward_std': '1.22753465', 'gathered_video_score2_mq_reward_mean': '3.09765625', 'gathered_video_score2_mq_reward_std': '0.36545888', 'gathered_video_score2_overall_reward_mean': '8.75146484', 'gathered_video_score2_overall_reward_std': '1.22471285', 'gathered_video_score2_ta_reward_mean': '3.01513672', 'gathered_video_score2_ta_reward_std': '0.50118852', 'gathered_video_score2_vq_reward_mean': '2.63745117', 'gathered_video_score2_vq_reward_std': '0.53502387', 'gathered_ori_avg_reward_mean': '8.75024414', 'gathered_ori_avg_reward_std': '1.22753477'} +2026-05-29 23:16:52.833 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 23:16:57.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_0/index_7446_seed_6033_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 23:18:20.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:20.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:35.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_0/index_7446_seed_6034_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:19:57.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 23:19:57.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:13.783 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 11 train_steps 440 {'step': '11/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.00635973', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '303.84472656', 'grpo_loop_time': '430.00271606', 'ref_model_time': '121.80107117', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86628354', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.23852539', 'token_credit_video_advantage_mean': '-6.286e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.486e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00852847', 'kl_moving': '0.00852847', 'kl_stepwise': '-1.975e-08', 'gathered_avg_reward_mean': '8.24157715', 'gathered_avg_reward_std': '1.21948552', 'gathered_video_score2_mq_reward_mean': '2.91918945', 'gathered_video_score2_mq_reward_std': '0.39461637', 'gathered_video_score2_overall_reward_mean': '8.23852539', 'gathered_video_score2_overall_reward_std': '1.22101164', 'gathered_video_score2_ta_reward_mean': '2.80383301', 'gathered_video_score2_ta_reward_std': '0.60222656', 'gathered_video_score2_vq_reward_mean': '2.51855469', 'gathered_video_score2_vq_reward_std': '0.53660929', 'gathered_ori_avg_reward_mean': '8.24157715', 'gathered_ori_avg_reward_std': '1.21948552'} +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 23:31:14.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_0/index_6954_seed_18434_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 23:32:36.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:36.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:33:53.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_0/index_6954_seed_18435_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:34:14.971 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 23:34:15.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:28.799 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 12 train_steps 480 {'step': '12/10000', 'learning_rate': '0.00001000', 'loss': '-1.164e-08', 'grad_norm': '0.00644782', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '303.70025635', 'grpo_loop_time': '430.87988281', 'ref_model_time': '120.94725800', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86968064', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.54687500', 'token_credit_video_advantage_mean': '-2.328e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.540e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.164e-08', 'kl_loss': '0.00768734', 'kl_moving': '0.00768734', 'kl_stepwise': '-2.003e-08', 'gathered_avg_reward_mean': '8.54687500', 'gathered_avg_reward_std': '1.44763064', 'gathered_video_score2_mq_reward_mean': '3.04223633', 'gathered_video_score2_mq_reward_std': '0.50787580', 'gathered_video_score2_overall_reward_mean': '8.54687500', 'gathered_video_score2_overall_reward_std': '1.44666326', 'gathered_video_score2_ta_reward_mean': '2.78479004', 'gathered_video_score2_ta_reward_std': '0.56572217', 'gathered_video_score2_vq_reward_mean': '2.71984863', 'gathered_video_score2_vq_reward_std': '0.65687150', 'gathered_ori_avg_reward_mean': '8.54687500', 'gathered_ori_avg_reward_std': '1.44763064'} +2026-05-29 23:45:29.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 23:45:29.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_0/index_728_seed_22358_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-29 23:46:51.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:51.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:09.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_0/index_728_seed_22359_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:48:30.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 23:48:31.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.594 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 13 train_steps 520 {'step': '13/10000', 'learning_rate': '0.00001000', 'loss': '-5.122e-09', 'grad_norm': '0.01877860', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '306.39837646', 'grpo_loop_time': '433.39376831', 'ref_model_time': '122.78451538', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86134827', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.55957031', 'token_credit_video_advantage_mean': '1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.480e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.122e-09', 'kl_loss': '0.00891272', 'kl_moving': '0.00891272', 'kl_stepwise': '-2.006e-08', 'gathered_avg_reward_mean': '8.55859375', 'gathered_avg_reward_std': '1.42015755', 'gathered_video_score2_mq_reward_mean': '2.96911621', 'gathered_video_score2_mq_reward_std': '0.45510209', 'gathered_video_score2_overall_reward_mean': '8.55957031', 'gathered_video_score2_overall_reward_std': '1.41579568', 'gathered_video_score2_ta_reward_mean': '2.98144531', 'gathered_video_score2_ta_reward_std': '0.76014775', 'gathered_video_score2_vq_reward_mean': '2.60803223', 'gathered_video_score2_vq_reward_std': '0.54641193', 'gathered_ori_avg_reward_mean': '8.55859375', 'gathered_ori_avg_reward_std': '1.42015755'} +2026-05-29 23:59:50.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 23:59:50.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:50.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:58.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_0/index_8365_seed_10793_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 00:01:15.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:01:15.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:26.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_0/index_8365_seed_10794_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:02:54.335 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 00:02:55.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:12.812 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 14 train_steps 560 {'step': '14/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00382404', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '309.40264893', 'grpo_loop_time': '430.20220947', 'ref_model_time': '123.57440186', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84791988', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91601562', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.425e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00680515', 'kl_moving': '0.00680515', 'kl_stepwise': '-1.838e-08', 'gathered_avg_reward_mean': '8.91564941', 'gathered_avg_reward_std': '1.28591383', 'gathered_video_score2_mq_reward_mean': '3.04125977', 'gathered_video_score2_mq_reward_std': '0.37168300', 'gathered_video_score2_overall_reward_mean': '8.91601562', 'gathered_video_score2_overall_reward_std': '1.28763974', 'gathered_video_score2_ta_reward_mean': '3.10522461', 'gathered_video_score2_ta_reward_std': '0.74501795', 'gathered_video_score2_vq_reward_mean': '2.76916504', 'gathered_video_score2_vq_reward_std': '0.48587480', 'gathered_ori_avg_reward_mean': '8.91564941', 'gathered_ori_avg_reward_std': '1.28591383'} +2026-05-30 00:14:13.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 00:14:13.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_0/index_4241_seed_33057_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 00:15:37.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:37.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:16:51.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_0/index_4241_seed_33058_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:17:17.275 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 00:17:18.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.518 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 15 train_steps 600 {'step': '15/10000', 'learning_rate': '0.00001000', 'loss': '8.382e-09', 'grad_norm': '0.00345457', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.44049072', 'grpo_loop_time': '433.07849121', 'ref_model_time': '121.17316437', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86010915', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.86425781', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.429e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.382e-09', 'kl_loss': '0.00677875', 'kl_moving': '0.00677875', 'kl_stepwise': '-1.999e-08', 'gathered_avg_reward_mean': '8.86364746', 'gathered_avg_reward_std': '1.00036228', 'gathered_video_score2_mq_reward_mean': '3.08911133', 'gathered_video_score2_mq_reward_std': '0.24222970', 'gathered_video_score2_overall_reward_mean': '8.86425781', 'gathered_video_score2_overall_reward_std': '1.00100994', 'gathered_video_score2_ta_reward_mean': '3.03857422', 'gathered_video_score2_ta_reward_std': '0.68330193', 'gathered_video_score2_vq_reward_mean': '2.73596191', 'gathered_video_score2_vq_reward_std': '0.44604897', 'gathered_ori_avg_reward_mean': '8.86364746', 'gathered_ori_avg_reward_std': '1.00036228'} +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 00:28:36.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_0/index_6121_seed_22864_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 00:30:03.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:30:03.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:16.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_0/index_6121_seed_22865_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:31:38.436 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 00:31:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.591 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 16 train_steps 640 {'step': '16/10000', 'learning_rate': '0.00001000', 'loss': '8.382e-09', 'grad_norm': '0.04185532', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '309.29934692', 'grpo_loop_time': '432.24121094', 'ref_model_time': '124.70022583', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88239175', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.87011719', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.337e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.382e-09', 'kl_loss': '0.00761216', 'kl_moving': '0.00761216', 'kl_stepwise': '-1.914e-08', 'gathered_avg_reward_mean': '8.87036133', 'gathered_avg_reward_std': '1.82110381', 'gathered_video_score2_mq_reward_mean': '3.08459473', 'gathered_video_score2_mq_reward_std': '0.62519616', 'gathered_video_score2_overall_reward_mean': '8.87011719', 'gathered_video_score2_overall_reward_std': '1.82114398', 'gathered_video_score2_ta_reward_mean': '3.05651855', 'gathered_video_score2_ta_reward_std': '0.71688038', 'gathered_video_score2_vq_reward_mean': '2.72924805', 'gathered_video_score2_vq_reward_std': '0.72108310', 'gathered_ori_avg_reward_mean': '8.87036133', 'gathered_ori_avg_reward_std': '1.82110381'} +2026-05-30 00:42:57.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 00:42:57.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_0/index_3814_seed_28667_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 00:44:20.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:20.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:45:38.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_0/index_3814_seed_28668_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:46:01.355 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 00:46:02.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.123 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 17 train_steps 680 {'step': '17/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00709957', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '306.89382935', 'grpo_loop_time': '430.34417725', 'ref_model_time': '121.05759430', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87071919', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.52197266', 'token_credit_video_advantage_mean': '-5.821e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.458e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.211e-08', 'kl_loss': '0.00728905', 'kl_moving': '0.00728906', 'kl_stepwise': '-1.829e-08', 'gathered_avg_reward_mean': '8.52050781', 'gathered_avg_reward_std': '0.92469442', 'gathered_video_score2_mq_reward_mean': '3.02978516', 'gathered_video_score2_mq_reward_std': '0.26608014', 'gathered_video_score2_overall_reward_mean': '8.52197266', 'gathered_video_score2_overall_reward_std': '0.93033850', 'gathered_video_score2_ta_reward_mean': '2.74914551', 'gathered_video_score2_ta_reward_std': '0.47274929', 'gathered_video_score2_vq_reward_mean': '2.74157715', 'gathered_video_score2_vq_reward_std': '0.47819725', 'gathered_ori_avg_reward_mean': '8.52050781', 'gathered_ori_avg_reward_std': '0.92469436'} +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 00:57:15.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_0/index_367_seed_17358_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 00:58:38.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:38.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:59:51.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_0/index_367_seed_17359_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:00:15.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 01:00:16.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:37.851 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 18 train_steps 720 {'step': '18/10000', 'learning_rate': '0.00001000', 'loss': '-2.142e-09', 'grad_norm': '0.00466378', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '305.51867676', 'grpo_loop_time': '432.68377686', 'ref_model_time': '123.03968048', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87310642', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.35595703', 'token_credit_video_advantage_mean': '-5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.443e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.142e-09', 'kl_loss': '0.00718543', 'kl_moving': '0.00718543', 'kl_stepwise': '-1.906e-08', 'gathered_avg_reward_mean': '8.35534668', 'gathered_avg_reward_std': '1.57904816', 'gathered_video_score2_mq_reward_mean': '3.04077148', 'gathered_video_score2_mq_reward_std': '0.52329624', 'gathered_video_score2_overall_reward_mean': '8.35595703', 'gathered_video_score2_overall_reward_std': '1.57696056', 'gathered_video_score2_ta_reward_mean': '2.81188965', 'gathered_video_score2_ta_reward_std': '0.76891172', 'gathered_video_score2_vq_reward_mean': '2.50268555', 'gathered_video_score2_vq_reward_std': '0.63301623', 'gathered_ori_avg_reward_mean': '8.35534668', 'gathered_ori_avg_reward_std': '1.57904816'} +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 01:11:38.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:45.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_0/index_8800_seed_13689_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 01:12:59.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:59.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:18.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_0/index_8800_seed_13690_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 01:14:42.923 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:14:43.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 01:14:44.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.370 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 19 train_steps 760 {'step': '19/10000', 'learning_rate': '0.00001000', 'loss': '6.263e-09', 'grad_norm': '0.00744222', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '309.06100464', 'grpo_loop_time': '432.32165527', 'ref_model_time': '121.82716370', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87860334', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.81201172', 'token_credit_video_advantage_mean': '1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.422e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.215e-09', 'kl_loss': '0.00735054', 'kl_moving': '0.00735054', 'kl_stepwise': '-2.009e-08', 'gathered_avg_reward_mean': '8.81372070', 'gathered_avg_reward_std': '1.27451718', 'gathered_video_score2_mq_reward_mean': '3.08801270', 'gathered_video_score2_mq_reward_std': '0.40228033', 'gathered_video_score2_overall_reward_mean': '8.81201172', 'gathered_video_score2_overall_reward_std': '1.27326369', 'gathered_video_score2_ta_reward_mean': '3.01867676', 'gathered_video_score2_ta_reward_std': '0.64048898', 'gathered_video_score2_vq_reward_mean': '2.70703125', 'gathered_video_score2_vq_reward_std': '0.52159959', 'gathered_ori_avg_reward_mean': '8.81372070', 'gathered_ori_avg_reward_std': '1.27451718'} +2026-05-30 01:26:05.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 01:26:05.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:14.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_0/index_657_seed_28953_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 01:27:30.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:30.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:28:42.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_0/index_657_seed_28954_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:29:03.569 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 01:29:04.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:40:22.570 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 20 train_steps 800 {'step': '20/10000', 'learning_rate': '0.00001000', 'loss': '-2.421e-08', 'grad_norm': '0.00513679', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '304.30441284', 'grpo_loop_time': '432.34960938', 'ref_model_time': '123.72409821', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87000954', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.70849609', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.479e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.421e-08', 'kl_loss': '0.00807026', 'kl_moving': '0.00807027', 'kl_stepwise': '-2.022e-08', 'gathered_avg_reward_mean': '8.70751953', 'gathered_avg_reward_std': '1.07856882', 'gathered_video_score2_mq_reward_mean': '3.08178711', 'gathered_video_score2_mq_reward_std': '0.37725905', 'gathered_video_score2_overall_reward_mean': '8.70849609', 'gathered_video_score2_overall_reward_std': '1.08110797', 'gathered_video_score2_ta_reward_mean': '3.03637695', 'gathered_video_score2_ta_reward_std': '0.48888871', 'gathered_video_score2_vq_reward_mean': '2.58935547', 'gathered_video_score2_vq_reward_std': '0.56830376', 'gathered_ori_avg_reward_mean': '8.70751953', 'gathered_ori_avg_reward_std': '1.07856882'} +2026-05-30 01:41:09.458 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 01:41:19.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_0/index_5415_seed_11619_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 01:42:41.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:41.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:43:56.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_0/index_5415_seed_11620_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:44:17.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 01:44:18.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.757 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 21 train_steps 840 {'step': '21/10000', 'learning_rate': '0.00001000', 'loss': '6.170e-09', 'grad_norm': '0.03727733', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.08618164', 'grpo_loop_time': '433.57009888', 'ref_model_time': '121.30229950', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85748756', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.68457031', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.444e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.307e-09', 'kl_loss': '0.00728368', 'kl_moving': '0.00728369', 'kl_stepwise': '-1.980e-08', 'gathered_avg_reward_mean': '8.68457031', 'gathered_avg_reward_std': '0.98758191', 'gathered_video_score2_mq_reward_mean': '3.05932617', 'gathered_video_score2_mq_reward_std': '0.24917741', 'gathered_video_score2_overall_reward_mean': '8.68457031', 'gathered_video_score2_overall_reward_std': '0.98706412', 'gathered_video_score2_ta_reward_mean': '3.00500488', 'gathered_video_score2_ta_reward_std': '0.57481861', 'gathered_video_score2_vq_reward_mean': '2.62023926', 'gathered_video_score2_vq_reward_std': '0.59090930', 'gathered_ori_avg_reward_mean': '8.68457031', 'gathered_ori_avg_reward_std': '0.98758191'} +2026-05-30 01:55:37.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 01:55:38.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:45.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_0/index_6892_seed_11576_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 01:57:02.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:57:02.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:14.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_0/index_6892_seed_11577_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 01:58:38.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:58:38.704 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 01:58:39.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.652 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 22 train_steps 880 {'step': '22/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-10', 'grad_norm': '0.00460153', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '306.85894775', 'grpo_loop_time': '432.31222534', 'ref_model_time': '123.79095459', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85696799', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.78564453', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.465e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-10', 'kl_loss': '0.00735025', 'kl_moving': '0.00735025', 'kl_stepwise': '-2.029e-08', 'gathered_avg_reward_mean': '8.78222656', 'gathered_avg_reward_std': '0.93918973', 'gathered_video_score2_mq_reward_mean': '3.09130859', 'gathered_video_score2_mq_reward_std': '0.28893340', 'gathered_video_score2_overall_reward_mean': '8.78564453', 'gathered_video_score2_overall_reward_std': '0.94427675', 'gathered_video_score2_ta_reward_mean': '2.97839355', 'gathered_video_score2_ta_reward_std': '0.57403398', 'gathered_video_score2_vq_reward_mean': '2.71252441', 'gathered_video_score2_vq_reward_std': '0.51857436', 'gathered_ori_avg_reward_mean': '8.78222656', 'gathered_ori_avg_reward_std': '0.93918967'} +2026-05-30 02:09:57.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 02:09:58.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:05.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_0/index_4192_seed_17251_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 02:11:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:18.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:12:39.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_0/index_4192_seed_17252_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:13:02.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 02:13:02.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:22.918 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 23 train_steps 920 {'step': '23/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00598443', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '310.33398438', 'grpo_loop_time': '432.14703369', 'ref_model_time': '123.89632416', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85141480', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23974609', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.395e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00712895', 'kl_moving': '0.00712896', 'kl_stepwise': '-1.951e-08', 'gathered_avg_reward_mean': '9.24243164', 'gathered_avg_reward_std': '1.22607720', 'gathered_video_score2_mq_reward_mean': '3.26049805', 'gathered_video_score2_mq_reward_std': '0.40084344', 'gathered_video_score2_overall_reward_mean': '9.23974609', 'gathered_video_score2_overall_reward_std': '1.22433436', 'gathered_video_score2_ta_reward_mean': '3.15319824', 'gathered_video_score2_ta_reward_std': '0.59543031', 'gathered_video_score2_vq_reward_mean': '2.82873535', 'gathered_video_score2_vq_reward_std': '0.56627375', 'gathered_ori_avg_reward_mean': '9.24243164', 'gathered_ori_avg_reward_std': '1.22607720'} +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 02:24:23.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_0/index_7474_seed_23209_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 02:25:50.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:50.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:05.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_0/index_7474_seed_23210_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:27:32.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 02:27:33.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.053 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 24 train_steps 960 {'step': '24/10000', 'learning_rate': '0.00001000', 'loss': '1.420e-08', 'grad_norm': '0.00477155', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '316.50558472', 'grpo_loop_time': '432.97335815', 'ref_model_time': '124.73083496', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88819444', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.83886719', 'token_credit_video_advantage_mean': '-1.327e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.421e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.001e-08', 'kl_loss': '0.00719495', 'kl_moving': '0.00719495', 'kl_stepwise': '-2.010e-08', 'gathered_avg_reward_mean': '8.83923340', 'gathered_avg_reward_std': '1.10720897', 'gathered_video_score2_mq_reward_mean': '3.08544922', 'gathered_video_score2_mq_reward_std': '0.34324992', 'gathered_video_score2_overall_reward_mean': '8.83886719', 'gathered_video_score2_overall_reward_std': '1.11145294', 'gathered_video_score2_ta_reward_mean': '2.94128418', 'gathered_video_score2_ta_reward_std': '0.54480988', 'gathered_video_score2_vq_reward_mean': '2.81250000', 'gathered_video_score2_vq_reward_std': '0.59578317', 'gathered_ori_avg_reward_mean': '8.83923340', 'gathered_ori_avg_reward_std': '1.10720897'} +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 02:38:56.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_0/index_9364_seed_12316_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 02:40:20.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:20.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_0/index_9364_seed_12317_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:41:57.228 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 02:41:58.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:53:16.807 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 25 train_steps 1000 {'step': '25/10000', 'learning_rate': '0.00001000', 'loss': '1.397e-08', 'grad_norm': '0.00447244', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '305.24438477', 'grpo_loop_time': '432.11392212', 'ref_model_time': '122.33375549', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84200567', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.59130859', 'token_credit_video_advantage_mean': '-1.228e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.386e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-08', 'kl_loss': '0.00733975', 'kl_moving': '0.00733976', 'kl_stepwise': '-1.922e-08', 'gathered_avg_reward_mean': '8.59179688', 'gathered_avg_reward_std': '0.85680544', 'gathered_video_score2_mq_reward_mean': '3.04589844', 'gathered_video_score2_mq_reward_std': '0.23794281', 'gathered_video_score2_overall_reward_mean': '8.59130859', 'gathered_video_score2_overall_reward_std': '0.85410857', 'gathered_video_score2_ta_reward_mean': '2.99548340', 'gathered_video_score2_ta_reward_std': '0.49854794', 'gathered_video_score2_vq_reward_mean': '2.55041504', 'gathered_video_score2_vq_reward_std': '0.48574916', 'gathered_ori_avg_reward_mean': '8.59179688', 'gathered_ori_avg_reward_std': '0.85680544'} +2026-05-30 02:54:31.152 | INFO | __main__:save_checkpoint:1917 - Timer checkpoint saved at step 25 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-25 +2026-05-30 02:54:31.291 | INFO | __main__:_write_timer_marker:1761 - Timer resume marker updated at step 25: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-25 +2026-05-30 02:54:31.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 02:54:31.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_0/index_1319_seed_27710_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 02:55:51.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:51.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:08.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_0/index_1319_seed_27711_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 02:57:30.214 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:57:30.333 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 02:57:31.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.283 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 26 train_steps 1040 {'step': '26/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00543381', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.80526733', 'grpo_loop_time': '433.34197998', 'ref_model_time': '121.81600189', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86042070', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.35644531', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.480e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00738046', 'kl_moving': '0.00738046', 'kl_stepwise': '-2.047e-08', 'gathered_avg_reward_mean': '8.35498047', 'gathered_avg_reward_std': '1.50681591', 'gathered_video_score2_mq_reward_mean': '2.95568848', 'gathered_video_score2_mq_reward_std': '0.55144584', 'gathered_video_score2_overall_reward_mean': '8.35644531', 'gathered_video_score2_overall_reward_std': '1.50810313', 'gathered_video_score2_ta_reward_mean': '2.75524902', 'gathered_video_score2_ta_reward_std': '0.62358093', 'gathered_video_score2_vq_reward_mean': '2.64404297', 'gathered_video_score2_vq_reward_std': '0.56958443', 'gathered_ori_avg_reward_mean': '8.35498047', 'gathered_ori_avg_reward_std': '1.50681591'} +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 03:08:50.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_0/index_3260_seed_679_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 03:10:11.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:10:11.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:26.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_0/index_3260_seed_680_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 03:11:52.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:11:52.687 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 03:11:53.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.363 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 27 train_steps 1080 {'step': '27/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00408357', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.23947144', 'grpo_loop_time': '430.04965210', 'ref_model_time': '122.67352295', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83920711', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.61425781', 'token_credit_video_advantage_mean': '-2.561e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.426e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00696002', 'kl_moving': '0.00696002', 'kl_stepwise': '-1.940e-08', 'gathered_avg_reward_mean': '8.61962891', 'gathered_avg_reward_std': '0.96204525', 'gathered_video_score2_mq_reward_mean': '3.06689453', 'gathered_video_score2_mq_reward_std': '0.32054427', 'gathered_video_score2_overall_reward_mean': '8.61425781', 'gathered_video_score2_overall_reward_std': '0.96162355', 'gathered_video_score2_ta_reward_mean': '2.84997559', 'gathered_video_score2_ta_reward_std': '0.60214591', 'gathered_video_score2_vq_reward_mean': '2.70275879', 'gathered_video_score2_vq_reward_std': '0.50264126', 'gathered_ori_avg_reward_mean': '8.61962891', 'gathered_ori_avg_reward_std': '0.96204531'} +2026-05-30 03:23:09.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 03:23:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:17.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_0/index_2884_seed_11115_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 03:24:32.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:32.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:25:46.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_0/index_2884_seed_11116_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 03:26:07.442 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:26:07.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 03:26:08.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.346 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 28 train_steps 1120 {'step': '28/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00586424', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.58078003', 'grpo_loop_time': '430.03881836', 'ref_model_time': '122.35002899', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89147055', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.83520508', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.457e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00868107', 'kl_moving': '0.00868107', 'kl_stepwise': '-1.883e-08', 'gathered_avg_reward_mean': '8.83374023', 'gathered_avg_reward_std': '1.50240421', 'gathered_video_score2_mq_reward_mean': '3.10473633', 'gathered_video_score2_mq_reward_std': '0.46124095', 'gathered_video_score2_overall_reward_mean': '8.83520508', 'gathered_video_score2_overall_reward_std': '1.50443614', 'gathered_video_score2_ta_reward_mean': '3.10998535', 'gathered_video_score2_ta_reward_std': '0.59151369', 'gathered_video_score2_vq_reward_mean': '2.61901855', 'gathered_video_score2_vq_reward_std': '0.70320469', 'gathered_ori_avg_reward_mean': '8.83374023', 'gathered_ori_avg_reward_std': '1.50240421'} +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 03:37:26.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_0/index_7718_seed_23896_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 03:38:47.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:47.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:07.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_0/index_7718_seed_23897_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 03:40:38.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:40:38.417 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 03:40:39.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:58.925 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 29 train_steps 1160 {'step': '29/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00678085', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '318.67013550', 'grpo_loop_time': '433.31814575', 'ref_model_time': '124.84017944', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85327619', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.94042969', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.00719910', 'kl_moving': '0.00719910', 'kl_stepwise': '-1.924e-08', 'gathered_avg_reward_mean': '8.94104004', 'gathered_avg_reward_std': '1.39080393', 'gathered_video_score2_mq_reward_mean': '3.16699219', 'gathered_video_score2_mq_reward_std': '0.46050590', 'gathered_video_score2_overall_reward_mean': '8.94042969', 'gathered_video_score2_overall_reward_std': '1.39572585', 'gathered_video_score2_ta_reward_mean': '3.02404785', 'gathered_video_score2_ta_reward_std': '0.55337465', 'gathered_video_score2_vq_reward_mean': '2.75000000', 'gathered_video_score2_vq_reward_std': '0.63164949', 'gathered_ori_avg_reward_mean': '8.94104004', 'gathered_ori_avg_reward_std': '1.39080393'} +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 03:51:59.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_0/index_6763_seed_10134_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 03:53:23.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:23.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:54:38.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_0/index_6763_seed_10135_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:55:03.558 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 03:55:04.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:06:25.204 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 30 train_steps 1200 {'step': '30/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.00502035', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '310.70303345', 'grpo_loop_time': '432.28283691', 'ref_model_time': '123.78269958', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.90137815', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01269531', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.422e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.794e-09', 'kl_loss': '0.00758711', 'kl_moving': '0.00758712', 'kl_stepwise': '-1.888e-08', 'gathered_avg_reward_mean': '9.01232910', 'gathered_avg_reward_std': '1.55010891', 'gathered_video_score2_mq_reward_mean': '3.14367676', 'gathered_video_score2_mq_reward_std': '0.52178538', 'gathered_video_score2_overall_reward_mean': '9.01269531', 'gathered_video_score2_overall_reward_std': '1.55535197', 'gathered_video_score2_ta_reward_mean': '2.88427734', 'gathered_video_score2_ta_reward_std': '0.70902187', 'gathered_video_score2_vq_reward_mean': '2.98437500', 'gathered_video_score2_vq_reward_std': '0.58639467', 'gathered_ori_avg_reward_mean': '9.01232910', 'gathered_ori_avg_reward_std': '1.55010891'} +2026-05-30 04:07:14.428 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 04:07:19.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:28.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_0/index_8075_seed_1736_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 04:08:40.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:40.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:01.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_0/index_8075_seed_1737_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:10:27.342 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 04:10:28.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.264 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 31 train_steps 1240 {'step': '31/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00375231', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '311.43673706', 'grpo_loop_time': '435.60736084', 'ref_model_time': '121.03488922', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87472993', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.73681641', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.428e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00670568', 'kl_moving': '0.00670568', 'kl_stepwise': '-2.032e-08', 'gathered_avg_reward_mean': '8.73620605', 'gathered_avg_reward_std': '1.10692465', 'gathered_video_score2_mq_reward_mean': '3.12194824', 'gathered_video_score2_mq_reward_std': '0.30912367', 'gathered_video_score2_overall_reward_mean': '8.73681641', 'gathered_video_score2_overall_reward_std': '1.10942733', 'gathered_video_score2_ta_reward_mean': '2.90808105', 'gathered_video_score2_ta_reward_std': '0.60564476', 'gathered_video_score2_vq_reward_mean': '2.70617676', 'gathered_video_score2_vq_reward_std': '0.55693114', 'gathered_ori_avg_reward_mean': '8.73620605', 'gathered_ori_avg_reward_std': '1.10692465'} +2026-05-30 04:21:48.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 04:21:48.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:59.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_0/index_9874_seed_23792_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 04:23:16.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:23:16.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:29.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_0/index_9874_seed_23793_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:24:53.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 04:24:54.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.369 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 32 train_steps 1280 {'step': '32/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00535139', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '309.59136963', 'grpo_loop_time': '433.30490112', 'ref_model_time': '122.24298096', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85538250', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.29101562', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.385e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00668656', 'kl_moving': '0.00668656', 'kl_stepwise': '-1.869e-08', 'gathered_avg_reward_mean': '9.29089355', 'gathered_avg_reward_std': '1.37223244', 'gathered_video_score2_mq_reward_mean': '3.20715332', 'gathered_video_score2_mq_reward_std': '0.46001795', 'gathered_video_score2_overall_reward_mean': '9.29101562', 'gathered_video_score2_overall_reward_std': '1.37069726', 'gathered_video_score2_ta_reward_mean': '3.20483398', 'gathered_video_score2_ta_reward_std': '0.65723670', 'gathered_video_score2_vq_reward_mean': '2.87890625', 'gathered_video_score2_vq_reward_std': '0.63178235', 'gathered_ori_avg_reward_mean': '9.29089355', 'gathered_ori_avg_reward_std': '1.37223244'} +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 04:36:12.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:20.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_0/index_5918_seed_27276_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 04:37:35.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:35.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:38:50.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_0/index_5918_seed_27277_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:39:15.477 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 04:39:16.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.279 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 33 train_steps 1320 {'step': '33/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00424584', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '310.04168701', 'grpo_loop_time': '435.03982544', 'ref_model_time': '125.00785828', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86717856', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24707031', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.427e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-09', 'kl_loss': '0.00684777', 'kl_moving': '0.00684777', 'kl_stepwise': '-1.945e-08', 'gathered_avg_reward_mean': '9.24975586', 'gathered_avg_reward_std': '1.30585575', 'gathered_video_score2_mq_reward_mean': '3.19433594', 'gathered_video_score2_mq_reward_std': '0.47200307', 'gathered_video_score2_overall_reward_mean': '9.24707031', 'gathered_video_score2_overall_reward_std': '1.30483627', 'gathered_video_score2_ta_reward_mean': '3.24011230', 'gathered_video_score2_ta_reward_std': '0.53943259', 'gathered_video_score2_vq_reward_mean': '2.81530762', 'gathered_video_score2_vq_reward_std': '0.60065246', 'gathered_ori_avg_reward_mean': '9.24975586', 'gathered_ori_avg_reward_std': '1.30585575'} +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 04:50:41.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:48.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_0/index_113_seed_5468_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 04:52:02.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:52:02.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:19.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_0/index_113_seed_5469_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 04:53:43.265 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:53:43.378 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 04:53:44.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.657 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 34 train_steps 1360 {'step': '34/10000', 'learning_rate': '0.00001000', 'loss': '3.492e-09', 'grad_norm': '0.00649257', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.91320801', 'grpo_loop_time': '430.90628052', 'ref_model_time': '123.84153748', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86291087', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.68847656', 'token_credit_video_advantage_mean': '-6.985e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.630e-09', 'kl_loss': '0.00760074', 'kl_moving': '0.00760074', 'kl_stepwise': '-1.937e-08', 'gathered_avg_reward_mean': '8.68713379', 'gathered_avg_reward_std': '1.19572449', 'gathered_video_score2_mq_reward_mean': '3.11193848', 'gathered_video_score2_mq_reward_std': '0.40013582', 'gathered_video_score2_overall_reward_mean': '8.68847656', 'gathered_video_score2_overall_reward_std': '1.19529939', 'gathered_video_score2_ta_reward_mean': '2.83105469', 'gathered_video_score2_ta_reward_std': '0.59541732', 'gathered_video_score2_vq_reward_mean': '2.74414062', 'gathered_video_score2_vq_reward_std': '0.60513157', 'gathered_ori_avg_reward_mean': '8.68713379', 'gathered_ori_avg_reward_std': '1.19572449'} +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 05:05:07.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_0/index_776_seed_23861_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 05:06:30.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:30.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:07:43.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_0/index_776_seed_23862_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 05:08:12.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:08:12.161 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 05:08:12.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.318 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 35 train_steps 1400 {'step': '35/10000', 'learning_rate': '0.00001000', 'loss': '-3.027e-09', 'grad_norm': '0.00514172', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '308.04806519', 'grpo_loop_time': '430.84201050', 'ref_model_time': '120.99861908', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85236770', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.99316406', 'token_credit_video_advantage_mean': '-5.821e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.488e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.027e-09', 'kl_loss': '0.00751917', 'kl_moving': '0.00751918', 'kl_stepwise': '-2.004e-08', 'gathered_avg_reward_mean': '8.99230957', 'gathered_avg_reward_std': '1.32947028', 'gathered_video_score2_mq_reward_mean': '3.13952637', 'gathered_video_score2_mq_reward_std': '0.38438836', 'gathered_video_score2_overall_reward_mean': '8.99316406', 'gathered_video_score2_overall_reward_std': '1.32862437', 'gathered_video_score2_ta_reward_mean': '3.01184082', 'gathered_video_score2_ta_reward_std': '0.75659966', 'gathered_video_score2_vq_reward_mean': '2.84094238', 'gathered_video_score2_vq_reward_std': '0.56146550', 'gathered_ori_avg_reward_mean': '8.99230957', 'gathered_ori_avg_reward_std': '1.32947028'} +2026-05-30 05:19:25.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 05:19:25.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_0/index_5179_seed_11323_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 05:20:46.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:46.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:21:58.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_0/index_5179_seed_11324_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:22:19.953 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 05:22:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.087 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 36 train_steps 1440 {'step': '36/10000', 'learning_rate': '0.00001000', 'loss': '-1.220e-08', 'grad_norm': '0.00364636', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '299.00164795', 'grpo_loop_time': '433.46362305', 'ref_model_time': '122.57185364', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88290977', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.60986328', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.428e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.220e-08', 'kl_loss': '0.00739299', 'kl_moving': '0.00739300', 'kl_stepwise': '-1.948e-08', 'gathered_avg_reward_mean': '8.61145020', 'gathered_avg_reward_std': '1.31141055', 'gathered_video_score2_mq_reward_mean': '3.01562500', 'gathered_video_score2_mq_reward_std': '0.39563438', 'gathered_video_score2_overall_reward_mean': '8.60986328', 'gathered_video_score2_overall_reward_std': '1.31007957', 'gathered_video_score2_ta_reward_mean': '2.93481445', 'gathered_video_score2_ta_reward_std': '0.61402601', 'gathered_video_score2_vq_reward_mean': '2.66101074', 'gathered_video_score2_vq_reward_std': '0.60015726', 'gathered_ori_avg_reward_mean': '8.61145020', 'gathered_ori_avg_reward_std': '1.31141055'} +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 05:33:38.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:45.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_0/index_5724_seed_12245_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 05:35:00.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:35:00.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:15.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_0/index_5724_seed_12246_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:36:36.897 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 05:36:37.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.129 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 37 train_steps 1480 {'step': '37/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00538062', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.41210938', 'grpo_loop_time': '434.86535645', 'ref_model_time': '121.78874969', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88884163', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.70068359', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.386e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00772136', 'kl_moving': '0.00772136', 'kl_stepwise': '-1.930e-08', 'gathered_avg_reward_mean': '8.70141602', 'gathered_avg_reward_std': '1.61267316', 'gathered_video_score2_mq_reward_mean': '3.06176758', 'gathered_video_score2_mq_reward_std': '0.59138536', 'gathered_video_score2_overall_reward_mean': '8.70068359', 'gathered_video_score2_overall_reward_std': '1.61360574', 'gathered_video_score2_ta_reward_mean': '3.05249023', 'gathered_video_score2_ta_reward_std': '0.62638009', 'gathered_video_score2_vq_reward_mean': '2.58715820', 'gathered_video_score2_vq_reward_std': '0.60443968', 'gathered_ori_avg_reward_mean': '8.70141602', 'gathered_ori_avg_reward_std': '1.61267316'} +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 05:47:59.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_0/index_6980_seed_10946_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 05:49:23.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:23.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:35.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_0/index_6980_seed_10947_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:50:57.695 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 05:50:58.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.751 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 38 train_steps 1520 {'step': '38/10000', 'learning_rate': '0.00001000', 'loss': '4.831e-09', 'grad_norm': '0.00425701', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '303.94531250', 'grpo_loop_time': '434.57275391', 'ref_model_time': '123.36450958', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83899128', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.74316406', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.455e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.969e-09', 'kl_loss': '0.00779265', 'kl_moving': '0.00779265', 'kl_stepwise': '-1.921e-08', 'gathered_avg_reward_mean': '8.74401855', 'gathered_avg_reward_std': '1.62021911', 'gathered_video_score2_mq_reward_mean': '3.05773926', 'gathered_video_score2_mq_reward_std': '0.48348942', 'gathered_video_score2_overall_reward_mean': '8.74316406', 'gathered_video_score2_overall_reward_std': '1.61990702', 'gathered_video_score2_ta_reward_mean': '2.95910645', 'gathered_video_score2_ta_reward_std': '0.80813098', 'gathered_video_score2_vq_reward_mean': '2.72717285', 'gathered_video_score2_vq_reward_std': '0.63298172', 'gathered_ori_avg_reward_mean': '8.74401855', 'gathered_ori_avg_reward_std': '1.62021911'} +2026-05-30 06:02:20.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 06:02:21.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:30.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_0/index_691_seed_10955_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 06:03:47.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:47.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:04.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_0/index_691_seed_10956_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:05:32.279 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 06:05:33.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.598 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 39 train_steps 1560 {'step': '39/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00467661', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '313.87023926', 'grpo_loop_time': '438.86425781', 'ref_model_time': '120.51562500', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85110509', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.77001953', 'token_credit_video_advantage_mean': '1.118e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.515e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.00758628', 'kl_moving': '0.00758628', 'kl_stepwise': '-2.020e-08', 'gathered_avg_reward_mean': '8.76989746', 'gathered_avg_reward_std': '1.42625880', 'gathered_video_score2_mq_reward_mean': '3.05749512', 'gathered_video_score2_mq_reward_std': '0.48587957', 'gathered_video_score2_overall_reward_mean': '8.77001953', 'gathered_video_score2_overall_reward_std': '1.42537308', 'gathered_video_score2_ta_reward_mean': '2.87890625', 'gathered_video_score2_ta_reward_std': '0.69117308', 'gathered_video_score2_vq_reward_mean': '2.83349609', 'gathered_video_score2_vq_reward_std': '0.54857761', 'gathered_ori_avg_reward_mean': '8.76989746', 'gathered_ori_avg_reward_std': '1.42625880'} +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 06:16:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:03.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_0/index_7124_seed_13630_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 06:18:17.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:17.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:31.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_0/index_7124_seed_13631_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 06:19:58.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:19:58.960 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 06:19:59.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:31:23.041 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 40 train_steps 1600 {'step': '40/10000', 'learning_rate': '0.00001000', 'loss': '1.304e-08', 'grad_norm': '0.00363743', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '310.51904297', 'grpo_loop_time': '435.42047119', 'ref_model_time': '124.12377930', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86974263', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.92236328', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.461e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.304e-08', 'kl_loss': '0.00714236', 'kl_moving': '0.00714237', 'kl_stepwise': '-1.993e-08', 'gathered_avg_reward_mean': '8.92028809', 'gathered_avg_reward_std': '0.99226075', 'gathered_video_score2_mq_reward_mean': '3.08349609', 'gathered_video_score2_mq_reward_std': '0.29642773', 'gathered_video_score2_overall_reward_mean': '8.92236328', 'gathered_video_score2_overall_reward_std': '0.99109489', 'gathered_video_score2_ta_reward_mean': '3.13122559', 'gathered_video_score2_ta_reward_std': '0.53927153', 'gathered_video_score2_vq_reward_mean': '2.70556641', 'gathered_video_score2_vq_reward_std': '0.45756486', 'gathered_ori_avg_reward_mean': '8.92028809', 'gathered_ori_avg_reward_std': '0.99226075'} +2026-05-30 06:32:12.186 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 06:32:17.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:26.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_0/index_4029_seed_10540_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 06:33:42.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:42.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:34:55.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_0/index_4029_seed_10541_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:35:15.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 06:35:16.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.566 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 41 train_steps 1640 {'step': '41/10000', 'learning_rate': '0.00001000', 'loss': '6.007e-09', 'grad_norm': '0.00470235', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '301.85211182', 'grpo_loop_time': '434.29522705', 'ref_model_time': '120.80400848', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87954921', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.70751953', 'token_credit_video_advantage_mean': '-5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.435e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.007e-09', 'kl_loss': '0.00736600', 'kl_moving': '0.00736600', 'kl_stepwise': '-1.854e-08', 'gathered_avg_reward_mean': '8.70825195', 'gathered_avg_reward_std': '0.87408417', 'gathered_video_score2_mq_reward_mean': '3.03686523', 'gathered_video_score2_mq_reward_std': '0.28832740', 'gathered_video_score2_overall_reward_mean': '8.70751953', 'gathered_video_score2_overall_reward_std': '0.87095129', 'gathered_video_score2_ta_reward_mean': '2.94165039', 'gathered_video_score2_ta_reward_std': '0.58118182', 'gathered_video_score2_vq_reward_mean': '2.72973633', 'gathered_video_score2_vq_reward_std': '0.42467719', 'gathered_ori_avg_reward_mean': '8.70825195', 'gathered_ori_avg_reward_std': '0.87408423'} +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 06:46:36.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:36.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_0/index_278_seed_9871_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 06:48:00.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:48:00.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:15.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_0/index_278_seed_9872_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:49:37.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 06:49:38.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:58.981 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 42 train_steps 1680 {'step': '42/10000', 'learning_rate': '0.00001000', 'loss': '-1.322e-08', 'grad_norm': '0.00425393', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.62261963', 'grpo_loop_time': '435.47766113', 'ref_model_time': '120.25177002', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86933297', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.92333984', 'token_credit_video_advantage_mean': '6.985e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.418e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.322e-08', 'kl_loss': '0.00754245', 'kl_moving': '0.00754246', 'kl_stepwise': '-1.954e-08', 'gathered_avg_reward_mean': '8.91967773', 'gathered_avg_reward_std': '1.20032883', 'gathered_video_score2_mq_reward_mean': '3.11962891', 'gathered_video_score2_mq_reward_std': '0.36482388', 'gathered_video_score2_overall_reward_mean': '8.92333984', 'gathered_video_score2_overall_reward_std': '1.20041966', 'gathered_video_score2_ta_reward_mean': '3.07055664', 'gathered_video_score2_ta_reward_std': '0.58995199', 'gathered_video_score2_vq_reward_mean': '2.72949219', 'gathered_video_score2_vq_reward_std': '0.58540583', 'gathered_ori_avg_reward_mean': '8.91967773', 'gathered_ori_avg_reward_std': '1.20032883'} +2026-05-30 07:00:59.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 07:00:59.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_0/index_7277_seed_21237_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 07:02:21.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:21.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:38.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_0/index_7277_seed_21238_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 07:03:59.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:04:00.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 07:04:00.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.423 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 43 train_steps 1720 {'step': '43/10000', 'learning_rate': '0.00001000', 'loss': '-8.382e-09', 'grad_norm': '0.01173138', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.17480469', 'grpo_loop_time': '433.59423828', 'ref_model_time': '124.08420563', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85163814', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07226562', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.483e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.00754138', 'kl_moving': '0.00754139', 'kl_stepwise': '-1.993e-08', 'gathered_avg_reward_mean': '9.07641602', 'gathered_avg_reward_std': '1.12421596', 'gathered_video_score2_mq_reward_mean': '3.14440918', 'gathered_video_score2_mq_reward_std': '0.39243960', 'gathered_video_score2_overall_reward_mean': '9.07226562', 'gathered_video_score2_overall_reward_std': '1.12373614', 'gathered_video_score2_ta_reward_mean': '3.15368652', 'gathered_video_score2_ta_reward_std': '0.46752170', 'gathered_video_score2_vq_reward_mean': '2.77832031', 'gathered_video_score2_vq_reward_std': '0.56455535', 'gathered_ori_avg_reward_mean': '9.07641602', 'gathered_ori_avg_reward_std': '1.12421596'} +2026-05-30 07:15:21.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 07:15:21.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:29.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_0/index_656_seed_24741_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 07:16:43.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:43.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:06.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_0/index_656_seed_24742_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:18:40.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 07:18:41.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.270 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 44 train_steps 1760 {'step': '44/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00350566', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '326.07305908', 'grpo_loop_time': '435.56503296', 'ref_model_time': '124.39446259', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87391484', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.11669922', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.453e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00690541', 'kl_moving': '0.00690542', 'kl_stepwise': '-1.897e-08', 'gathered_avg_reward_mean': '9.11889648', 'gathered_avg_reward_std': '0.90126657', 'gathered_video_score2_mq_reward_mean': '3.10058594', 'gathered_video_score2_mq_reward_std': '0.37274382', 'gathered_video_score2_overall_reward_mean': '9.11669922', 'gathered_video_score2_overall_reward_std': '0.90157813', 'gathered_video_score2_ta_reward_mean': '3.04943848', 'gathered_video_score2_ta_reward_std': '0.49188718', 'gathered_video_score2_vq_reward_mean': '2.96887207', 'gathered_video_score2_vq_reward_std': '0.43771899', 'gathered_ori_avg_reward_mean': '9.11889648', 'gathered_ori_avg_reward_std': '0.90126663'} +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 07:30:08.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_0/index_7316_seed_27300_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 07:31:32.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:32.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:32:48.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_0/index_7316_seed_27301_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 07:33:10.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:33:10.721 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 07:33:11.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.299 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 45 train_steps 1800 {'step': '45/10000', 'learning_rate': '0.00001000', 'loss': '6.519e-09', 'grad_norm': '0.00604155', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '305.65631104', 'grpo_loop_time': '433.63363647', 'ref_model_time': '121.09364319', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87967372', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.88916016', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00796550', 'kl_moving': '0.00796550', 'kl_stepwise': '-1.975e-08', 'gathered_avg_reward_mean': '8.88879395', 'gathered_avg_reward_std': '1.47059274', 'gathered_video_score2_mq_reward_mean': '3.02478027', 'gathered_video_score2_mq_reward_std': '0.46781841', 'gathered_video_score2_overall_reward_mean': '8.88916016', 'gathered_video_score2_overall_reward_std': '1.46961129', 'gathered_video_score2_ta_reward_mean': '3.18579102', 'gathered_video_score2_ta_reward_std': '0.65284717', 'gathered_video_score2_vq_reward_mean': '2.67822266', 'gathered_video_score2_vq_reward_std': '0.61647922', 'gathered_ori_avg_reward_mean': '8.88879395', 'gathered_ori_avg_reward_std': '1.47059286'} +2026-05-30 07:44:30.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 07:44:30.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_0/index_7297_seed_12289_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 07:45:53.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:53.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:04.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_0/index_7297_seed_12290_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:47:29.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 07:47:30.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:58:59.786 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 46 train_steps 1840 {'step': '46/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.00505833', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '303.74484253', 'grpo_loop_time': '440.52148438', 'ref_model_time': '122.33473969', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87521029', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.90185547', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.415e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.794e-09', 'kl_loss': '0.00799767', 'kl_moving': '0.00799767', 'kl_stepwise': '-1.913e-08', 'gathered_avg_reward_mean': '8.90380859', 'gathered_avg_reward_std': '1.38339221', 'gathered_video_score2_mq_reward_mean': '3.12109375', 'gathered_video_score2_mq_reward_std': '0.45670351', 'gathered_video_score2_overall_reward_mean': '8.90185547', 'gathered_video_score2_overall_reward_std': '1.38352811', 'gathered_video_score2_ta_reward_mean': '3.05346680', 'gathered_video_score2_ta_reward_std': '0.57050514', 'gathered_video_score2_vq_reward_mean': '2.72924805', 'gathered_video_score2_vq_reward_std': '0.62949467', 'gathered_ori_avg_reward_mean': '8.90380859', 'gathered_ori_avg_reward_std': '1.38339233'} +2026-05-30 07:59:00.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 07:59:00.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_0/index_5451_seed_5374_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 08:00:20.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:20.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:01:41.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_0/index_5451_seed_5375_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 08:02:07.901 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:02:08.009 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 08:02:08.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.574 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 47 train_steps 1880 {'step': '47/10000', 'learning_rate': '0.00001000', 'loss': '1.118e-08', 'grad_norm': '0.00418627', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '313.16802979', 'grpo_loop_time': '436.97250366', 'ref_model_time': '123.10526276', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86371458', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23437500', 'token_credit_video_advantage_mean': '2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.417e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.118e-08', 'kl_loss': '0.00751305', 'kl_moving': '0.00751305', 'kl_stepwise': '-1.908e-08', 'gathered_avg_reward_mean': '9.23095703', 'gathered_avg_reward_std': '1.55898762', 'gathered_video_score2_mq_reward_mean': '3.17456055', 'gathered_video_score2_mq_reward_std': '0.49212012', 'gathered_video_score2_overall_reward_mean': '9.23437500', 'gathered_video_score2_overall_reward_std': '1.55802095', 'gathered_video_score2_ta_reward_mean': '3.21923828', 'gathered_video_score2_ta_reward_std': '0.63492316', 'gathered_video_score2_vq_reward_mean': '2.83715820', 'gathered_video_score2_vq_reward_std': '0.56656992', 'gathered_ori_avg_reward_mean': '9.23095703', 'gathered_ori_avg_reward_std': '1.55898762'} +2026-05-30 08:13:33.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 08:13:33.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:33.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_0/index_841_seed_31574_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 08:14:58.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:58.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:17.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_0/index_841_seed_31575_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:16:43.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 08:16:44.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:14.979 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 48 train_steps 1920 {'step': '48/10000', 'learning_rate': '0.00001000', 'loss': '-1.490e-08', 'grad_norm': '0.00506273', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '315.86846924', 'grpo_loop_time': '437.13812256', 'ref_model_time': '123.99699402', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86504614', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24560547', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.367e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.490e-08', 'kl_loss': '0.00728258', 'kl_moving': '0.00728258', 'kl_stepwise': '-1.794e-08', 'gathered_avg_reward_mean': '9.24548340', 'gathered_avg_reward_std': '1.34181452', 'gathered_video_score2_mq_reward_mean': '3.21166992', 'gathered_video_score2_mq_reward_std': '0.38374737', 'gathered_video_score2_overall_reward_mean': '9.24560547', 'gathered_video_score2_overall_reward_std': '1.34008706', 'gathered_video_score2_ta_reward_mean': '3.02526855', 'gathered_video_score2_ta_reward_std': '0.69798571', 'gathered_video_score2_vq_reward_mean': '3.00854492', 'gathered_video_score2_vq_reward_std': '0.60115606', 'gathered_ori_avg_reward_mean': '9.24548340', 'gathered_ori_avg_reward_std': '1.34181452'} +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 08:28:15.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_0/index_2883_seed_31728_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 08:29:39.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:39.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:30:55.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_0/index_2883_seed_31729_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:31:22.825 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 08:31:23.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.744 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 49 train_steps 1960 {'step': '49/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00518605', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '313.60449219', 'grpo_loop_time': '440.02203369', 'ref_model_time': '123.63328552', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89147639', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.43603516', 'token_credit_video_advantage_mean': '5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.553e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00794910', 'kl_moving': '0.00794910', 'kl_stepwise': '-2.077e-08', 'gathered_avg_reward_mean': '9.43664551', 'gathered_avg_reward_std': '1.40205133', 'gathered_video_score2_mq_reward_mean': '3.19653320', 'gathered_video_score2_mq_reward_std': '0.48463330', 'gathered_video_score2_overall_reward_mean': '9.43603516', 'gathered_video_score2_overall_reward_std': '1.39917302', 'gathered_video_score2_ta_reward_mean': '3.31762695', 'gathered_video_score2_ta_reward_std': '0.61487186', 'gathered_video_score2_vq_reward_mean': '2.92248535', 'gathered_video_score2_vq_reward_std': '0.55309021', 'gathered_ori_avg_reward_mean': '9.43664551', 'gathered_ori_avg_reward_std': '1.40205133'} +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 08:42:53.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_0/index_9953_seed_18575_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 08:44:20.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:20.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:45:35.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_0/index_9953_seed_18576_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:46:00.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 08:46:01.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:57:23.960 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 50 train_steps 2000 {'step': '50/10000', 'learning_rate': '0.00001000', 'loss': '3.260e-09', 'grad_norm': '0.00540651', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '314.12506104', 'grpo_loop_time': '436.14627075', 'ref_model_time': '123.96901703', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86075115', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.48974609', 'token_credit_video_advantage_mean': '5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.412e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.260e-09', 'kl_loss': '0.00866628', 'kl_moving': '0.00866628', 'kl_stepwise': '-1.973e-08', 'gathered_avg_reward_mean': '8.48571777', 'gathered_avg_reward_std': '1.01944566', 'gathered_video_score2_mq_reward_mean': '2.95117188', 'gathered_video_score2_mq_reward_std': '0.29700667', 'gathered_video_score2_overall_reward_mean': '8.48974609', 'gathered_video_score2_overall_reward_std': '1.02115464', 'gathered_video_score2_ta_reward_mean': '2.88085938', 'gathered_video_score2_ta_reward_std': '0.66999733', 'gathered_video_score2_vq_reward_mean': '2.65368652', 'gathered_video_score2_vq_reward_std': '0.49725169', 'gathered_ori_avg_reward_mean': '8.48571777', 'gathered_ori_avg_reward_std': '1.01944566'} +2026-05-30 08:58:31.940 | INFO | __main__:save_checkpoint:1917 - Timer checkpoint saved at step 50 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-50 +2026-05-30 08:58:32.083 | INFO | __main__:_write_timer_marker:1761 - Timer resume marker updated at step 50: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-50 +2026-05-30 08:58:34.877 | INFO | __main__:_prune_old_timer_checkpoints:1776 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-25 +2026-05-30 08:59:25.284 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 08:59:29.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:40.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_0/index_8040_seed_7961_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 09:00:54.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:54.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:11.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_0/index_8040_seed_7962_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:02:32.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 09:02:33.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.522 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 51 train_steps 2040 {'step': '51/10000', 'learning_rate': '0.00001000', 'loss': '1.024e-08', 'grad_norm': '0.00692541', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '309.29327393', 'grpo_loop_time': '439.98757935', 'ref_model_time': '124.13204193', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85004061', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.98876953', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.405e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00773189', 'kl_moving': '0.00773190', 'kl_stepwise': '-1.846e-08', 'gathered_avg_reward_mean': '8.99023438', 'gathered_avg_reward_std': '0.99922627', 'gathered_video_score2_mq_reward_mean': '3.05029297', 'gathered_video_score2_mq_reward_std': '0.26316458', 'gathered_video_score2_overall_reward_mean': '8.98876953', 'gathered_video_score2_overall_reward_std': '0.99983776', 'gathered_video_score2_ta_reward_mean': '3.23815918', 'gathered_video_score2_ta_reward_std': '0.56306767', 'gathered_video_score2_vq_reward_mean': '2.70178223', 'gathered_video_score2_vq_reward_std': '0.45462850', 'gathered_ori_avg_reward_mean': '8.99023438', 'gathered_ori_avg_reward_std': '0.99922627'} +2026-05-30 09:14:04.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 09:14:04.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:04.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:12.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_0/index_7704_seed_32713_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 09:15:28.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:28.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:16:52.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_0/index_7704_seed_32714_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:17:16.128 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 09:17:17.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.427 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 52 train_steps 2080 {'step': '52/10000', 'learning_rate': '0.00001000', 'loss': '9.127e-09', 'grad_norm': '0.00677766', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '314.68173218', 'grpo_loop_time': '437.65402222', 'ref_model_time': '121.26171875', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84243500', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.96191406', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.335e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.264e-09', 'kl_loss': '0.00693203', 'kl_moving': '0.00693203', 'kl_stepwise': '-1.918e-08', 'gathered_avg_reward_mean': '8.95874023', 'gathered_avg_reward_std': '1.10587287', 'gathered_video_score2_mq_reward_mean': '3.13452148', 'gathered_video_score2_mq_reward_std': '0.36879033', 'gathered_video_score2_overall_reward_mean': '8.96191406', 'gathered_video_score2_overall_reward_std': '1.11041260', 'gathered_video_score2_ta_reward_mean': '2.93359375', 'gathered_video_score2_ta_reward_std': '0.57222062', 'gathered_video_score2_vq_reward_mean': '2.89062500', 'gathered_video_score2_vq_reward_std': '0.62540269', 'gathered_ori_avg_reward_mean': '8.95874023', 'gathered_ori_avg_reward_std': '1.10587287'} +2026-05-30 09:28:42.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 09:28:42.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_0/index_5842_seed_17139_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 09:30:06.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:30:06.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:27.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_0/index_5842_seed_17140_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:31:53.234 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 09:31:54.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.000 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 53 train_steps 2120 {'step': '53/10000', 'learning_rate': '0.00001000', 'loss': '5.821e-10', 'grad_norm': '0.01244964', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '318.06481934', 'grpo_loop_time': '432.04806519', 'ref_model_time': '124.87544250', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87303561', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.70166016', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.470e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.821e-10', 'kl_loss': '0.00696919', 'kl_moving': '0.00696919', 'kl_stepwise': '-1.920e-08', 'gathered_avg_reward_mean': '8.69836426', 'gathered_avg_reward_std': '1.27811146', 'gathered_video_score2_mq_reward_mean': '3.04382324', 'gathered_video_score2_mq_reward_std': '0.45632252', 'gathered_video_score2_overall_reward_mean': '8.70166016', 'gathered_video_score2_overall_reward_std': '1.27935421', 'gathered_video_score2_ta_reward_mean': '2.92932129', 'gathered_video_score2_ta_reward_std': '0.60143638', 'gathered_video_score2_vq_reward_mean': '2.72521973', 'gathered_video_score2_vq_reward_std': '0.47833335', 'gathered_ori_avg_reward_mean': '8.69836426', 'gathered_ori_avg_reward_std': '1.27811146'} +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 09:43:17.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_0/index_608_seed_28453_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 09:44:41.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:41.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:02.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_0/index_608_seed_28454_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:46:24.133 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 09:46:25.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:45.724 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 54 train_steps 2160 {'step': '54/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00548060', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '309.68652344', 'grpo_loop_time': '435.84429932', 'ref_model_time': '120.76171112', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85082287', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91552734', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.448e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00774553', 'kl_moving': '0.00774553', 'kl_stepwise': '-2.015e-08', 'gathered_avg_reward_mean': '8.91796875', 'gathered_avg_reward_std': '1.03360820', 'gathered_video_score2_mq_reward_mean': '3.07775879', 'gathered_video_score2_mq_reward_std': '0.36954454', 'gathered_video_score2_overall_reward_mean': '8.91552734', 'gathered_video_score2_overall_reward_std': '1.03758824', 'gathered_video_score2_ta_reward_mean': '3.03955078', 'gathered_video_score2_ta_reward_std': '0.56815612', 'gathered_video_score2_vq_reward_mean': '2.80065918', 'gathered_video_score2_vq_reward_std': '0.50594610', 'gathered_ori_avg_reward_mean': '8.91796875', 'gathered_ori_avg_reward_std': '1.03360820'} +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 09:57:46.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:54.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_0/index_731_seed_19095_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 09:59:07.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:59:07.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:24.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_0/index_731_seed_19096_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:00:51.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 10:00:52.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.281 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 55 train_steps 2200 {'step': '55/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.01011332', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.35125732', 'grpo_loop_time': '431.93072510', 'ref_model_time': '119.83133698', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85023332', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.19287109', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.404e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00828453', 'kl_moving': '0.00828454', 'kl_stepwise': '-1.906e-08', 'gathered_avg_reward_mean': '8.19396973', 'gathered_avg_reward_std': '1.43067610', 'gathered_video_score2_mq_reward_mean': '2.91503906', 'gathered_video_score2_mq_reward_std': '0.45850727', 'gathered_video_score2_overall_reward_mean': '8.19287109', 'gathered_video_score2_overall_reward_std': '1.43098426', 'gathered_video_score2_ta_reward_mean': '2.75390625', 'gathered_video_score2_ta_reward_std': '0.63227427', 'gathered_video_score2_vq_reward_mean': '2.52502441', 'gathered_video_score2_vq_reward_std': '0.61670393', 'gathered_ori_avg_reward_mean': '8.19396973', 'gathered_ori_avg_reward_std': '1.43067610'} +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 10:12:07.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:16.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_0/index_4724_seed_8573_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 10:13:36.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:36.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:14:44.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_0/index_4724_seed_8574_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 10:15:05.855 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:15:05.965 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 10:15:06.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:23.855 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 56 train_steps 2240 {'step': '56/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00733044', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.21356201', 'grpo_loop_time': '432.54052734', 'ref_model_time': '121.52303314', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87590098', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.21337891', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.365e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00796513', 'kl_moving': '0.00796513', 'kl_stepwise': '-1.837e-08', 'gathered_avg_reward_mean': '9.21533203', 'gathered_avg_reward_std': '0.82602286', 'gathered_video_score2_mq_reward_mean': '3.13842773', 'gathered_video_score2_mq_reward_std': '0.22050197', 'gathered_video_score2_overall_reward_mean': '9.21337891', 'gathered_video_score2_overall_reward_std': '0.82538527', 'gathered_video_score2_ta_reward_mean': '3.14306641', 'gathered_video_score2_ta_reward_std': '0.61492097', 'gathered_video_score2_vq_reward_mean': '2.93383789', 'gathered_video_score2_vq_reward_std': '0.36286879', 'gathered_ori_avg_reward_mean': '9.21533203', 'gathered_ori_avg_reward_std': '0.82602280'} +2026-05-30 10:26:24.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 10:26:24.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_0/index_8869_seed_19024_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 10:27:45.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:02.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_0/index_8869_seed_19025_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:29:28.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 10:29:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:45.884 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 57 train_steps 2280 {'step': '57/10000', 'learning_rate': '0.00001000', 'loss': '-8.382e-09', 'grad_norm': '0.00500724', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '303.56201172', 'grpo_loop_time': '434.88845825', 'ref_model_time': '117.10438538', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84597290', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91845703', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.446e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.00787356', 'kl_moving': '0.00787356', 'kl_stepwise': '-1.979e-08', 'gathered_avg_reward_mean': '8.91638184', 'gathered_avg_reward_std': '1.27635324', 'gathered_video_score2_mq_reward_mean': '3.07751465', 'gathered_video_score2_mq_reward_std': '0.44446856', 'gathered_video_score2_overall_reward_mean': '8.91845703', 'gathered_video_score2_overall_reward_std': '1.27623367', 'gathered_video_score2_ta_reward_mean': '3.02185059', 'gathered_video_score2_ta_reward_std': '0.65681970', 'gathered_video_score2_vq_reward_mean': '2.81701660', 'gathered_video_score2_vq_reward_std': '0.51436359', 'gathered_ori_avg_reward_mean': '8.91638184', 'gathered_ori_avg_reward_std': '1.27635312'} +2026-05-30 10:40:46.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 10:40:46.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_0/index_7324_seed_9609_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 10:42:09.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:42:09.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:28.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_0/index_7324_seed_9610_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:43:54.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 10:43:55.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.423 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.301 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 58 train_steps 2320 {'step': '58/10000', 'learning_rate': '0.00001000', 'loss': '6.985e-09', 'grad_norm': '0.00406478', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '309.33593750', 'grpo_loop_time': '431.42575073', 'ref_model_time': '119.07804871', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85780048', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.26660156', 'token_credit_video_advantage_mean': '3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.454e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.985e-09', 'kl_loss': '0.00667232', 'kl_moving': '0.00667232', 'kl_stepwise': '-1.871e-08', 'gathered_avg_reward_mean': '9.26440430', 'gathered_avg_reward_std': '0.97992021', 'gathered_video_score2_mq_reward_mean': '3.20605469', 'gathered_video_score2_mq_reward_std': '0.31673011', 'gathered_video_score2_overall_reward_mean': '9.26660156', 'gathered_video_score2_overall_reward_std': '0.98412645', 'gathered_video_score2_ta_reward_mean': '2.99792480', 'gathered_video_score2_ta_reward_std': '0.56555742', 'gathered_video_score2_vq_reward_mean': '3.06042480', 'gathered_video_score2_vq_reward_std': '0.44859892', 'gathered_ori_avg_reward_mean': '9.26440430', 'gathered_ori_avg_reward_std': '0.97992021'} +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 10:55:08.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_0/index_4989_seed_16546_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 10:56:32.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:32.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:57:52.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_0/index_4989_seed_16547_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 10:58:22.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:58:22.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 10:58:23.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:39.734 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 59 train_steps 2360 {'step': '59/10000', 'learning_rate': '0.00001000', 'loss': '6.519e-09', 'grad_norm': '0.00584135', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '314.25759888', 'grpo_loop_time': '435.47439575', 'ref_model_time': '118.15214539', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88662952', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.40869141', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00807210', 'kl_moving': '0.00807210', 'kl_stepwise': '-2.046e-08', 'gathered_avg_reward_mean': '8.41333008', 'gathered_avg_reward_std': '0.90911907', 'gathered_video_score2_mq_reward_mean': '2.99572754', 'gathered_video_score2_mq_reward_std': '0.29815376', 'gathered_video_score2_overall_reward_mean': '8.40869141', 'gathered_video_score2_overall_reward_std': '0.90980083', 'gathered_video_score2_ta_reward_mean': '2.83703613', 'gathered_video_score2_ta_reward_std': '0.52033627', 'gathered_video_score2_vq_reward_mean': '2.58056641', 'gathered_video_score2_vq_reward_std': '0.49918082', 'gathered_ori_avg_reward_mean': '8.41333008', 'gathered_ori_avg_reward_std': '0.90911907'} +2026-05-30 11:09:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 11:09:40.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_0/index_4108_seed_23778_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 11:11:04.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:11:04.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:20.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_0/index_4108_seed_23779_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 11:12:42.201 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:12:42.332 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 11:12:43.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:24:08.702 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 60 train_steps 2400 {'step': '60/10000', 'learning_rate': '0.00001000', 'loss': '1.118e-08', 'grad_norm': '0.00540602', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.36404419', 'grpo_loop_time': '435.53198242', 'ref_model_time': '122.70092010', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86869961', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.63574219', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.454e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00819605', 'kl_moving': '0.00819605', 'kl_stepwise': '-1.900e-08', 'gathered_avg_reward_mean': '8.63623047', 'gathered_avg_reward_std': '1.14542484', 'gathered_video_score2_mq_reward_mean': '2.97985840', 'gathered_video_score2_mq_reward_std': '0.36327022', 'gathered_video_score2_overall_reward_mean': '8.63574219', 'gathered_video_score2_overall_reward_std': '1.14444757', 'gathered_video_score2_ta_reward_mean': '3.11499023', 'gathered_video_score2_ta_reward_std': '0.56123799', 'gathered_video_score2_vq_reward_mean': '2.54138184', 'gathered_video_score2_vq_reward_std': '0.53181100', 'gathered_ori_avg_reward_mean': '8.63623047', 'gathered_ori_avg_reward_std': '1.14542484'} +2026-05-30 11:25:03.237 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 11:25:07.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_0/index_7288_seed_20155_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 11:26:35.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:35.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:27:53.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_0/index_7288_seed_20156_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 11:28:20.076 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:28:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 11:28:21.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.306 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 61 train_steps 2440 {'step': '61/10000', 'learning_rate': '0.00001000', 'loss': '7.101e-09', 'grad_norm': '0.00668377', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '315.18856812', 'grpo_loop_time': '436.64425659', 'ref_model_time': '120.68167114', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88955617', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91503906', 'token_credit_video_advantage_mean': '-1.490e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.101e-09', 'kl_loss': '0.00815578', 'kl_moving': '0.00815579', 'kl_stepwise': '-1.975e-08', 'gathered_avg_reward_mean': '8.92065430', 'gathered_avg_reward_std': '1.08721554', 'gathered_video_score2_mq_reward_mean': '3.09973145', 'gathered_video_score2_mq_reward_std': '0.37301552', 'gathered_video_score2_overall_reward_mean': '8.91503906', 'gathered_video_score2_overall_reward_std': '1.08680081', 'gathered_video_score2_ta_reward_mean': '3.19775391', 'gathered_video_score2_ta_reward_std': '0.42482349', 'gathered_video_score2_vq_reward_mean': '2.62316895', 'gathered_video_score2_vq_reward_std': '0.52930903', 'gathered_ori_avg_reward_mean': '8.92065430', 'gathered_ori_avg_reward_std': '1.08721566'} +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 11:39:42.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_0/index_7919_seed_4851_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 11:41:03.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:41:03.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:19.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_0/index_7919_seed_4852_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 11:42:41.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:42:41.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 11:42:42.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:05.738 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 62 train_steps 2480 {'step': '62/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00460264', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.94949341', 'grpo_loop_time': '437.30825806', 'ref_model_time': '121.62802887', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87198168', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91894531', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.473e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00767305', 'kl_moving': '0.00767305', 'kl_stepwise': '-1.933e-08', 'gathered_avg_reward_mean': '8.91845703', 'gathered_avg_reward_std': '0.94080013', 'gathered_video_score2_mq_reward_mean': '3.07238770', 'gathered_video_score2_mq_reward_std': '0.31902143', 'gathered_video_score2_overall_reward_mean': '8.91894531', 'gathered_video_score2_overall_reward_std': '0.94333661', 'gathered_video_score2_ta_reward_mean': '3.08203125', 'gathered_video_score2_ta_reward_std': '0.60846817', 'gathered_video_score2_vq_reward_mean': '2.76403809', 'gathered_video_score2_vq_reward_std': '0.43144435', 'gathered_ori_avg_reward_mean': '8.91845703', 'gathered_ori_avg_reward_std': '0.94080013'} +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 11:54:06.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:14.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_0/index_2641_seed_15967_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 11:55:29.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:29.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:56:42.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_0/index_2641_seed_15968_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:57:05.105 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 11:57:06.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.063 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 63 train_steps 2520 {'step': '63/10000', 'learning_rate': '0.00001000', 'loss': '1.304e-08', 'grad_norm': '0.00504290', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '308.21841431', 'grpo_loop_time': '433.62020874', 'ref_model_time': '126.57654572', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88920462', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.92431641', 'token_credit_video_advantage_mean': '-7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.476e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00738452', 'kl_moving': '0.00738452', 'kl_stepwise': '-1.981e-08', 'gathered_avg_reward_mean': '8.92504883', 'gathered_avg_reward_std': '1.11535048', 'gathered_video_score2_mq_reward_mean': '3.17712402', 'gathered_video_score2_mq_reward_std': '0.40630177', 'gathered_video_score2_overall_reward_mean': '8.92431641', 'gathered_video_score2_overall_reward_std': '1.11976326', 'gathered_video_score2_ta_reward_mean': '2.88391113', 'gathered_video_score2_ta_reward_std': '0.57420689', 'gathered_video_score2_vq_reward_mean': '2.86401367', 'gathered_video_score2_vq_reward_std': '0.61115605', 'gathered_ori_avg_reward_mean': '8.92504883', 'gathered_ori_avg_reward_std': '1.11535048'} +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 12:08:30.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_0/index_2475_seed_8902_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 12:09:52.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:52.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:07.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_0/index_2475_seed_8903_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 12:11:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:11:29.132 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 12:11:30.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.031 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 64 train_steps 2560 {'step': '64/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00465028', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.28137207', 'grpo_loop_time': '438.19992065', 'ref_model_time': '121.10561371', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87730759', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.69360352', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.440e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00772239', 'kl_moving': '0.00772239', 'kl_stepwise': '-1.971e-08', 'gathered_avg_reward_mean': '8.69104004', 'gathered_avg_reward_std': '1.41845024', 'gathered_video_score2_mq_reward_mean': '2.96630859', 'gathered_video_score2_mq_reward_std': '0.47527578', 'gathered_video_score2_overall_reward_mean': '8.69360352', 'gathered_video_score2_overall_reward_std': '1.41893053', 'gathered_video_score2_ta_reward_mean': '3.02563477', 'gathered_video_score2_ta_reward_std': '0.64362872', 'gathered_video_score2_vq_reward_mean': '2.69909668', 'gathered_video_score2_vq_reward_std': '0.55658066', 'gathered_ori_avg_reward_mean': '8.69104004', 'gathered_ori_avg_reward_std': '1.41845024'} +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 12:22:53.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_0/index_5671_seed_5319_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 12:24:22.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:22.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:30.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_0/index_5671_seed_5320_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:25:54.727 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 12:25:55.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.655 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 65 train_steps 2600 {'step': '65/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00428361', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '302.97726440', 'grpo_loop_time': '434.30645752', 'ref_model_time': '119.52919006', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85250211', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20800781', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.477e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00711911', 'kl_moving': '0.00711911', 'kl_stepwise': '-1.976e-08', 'gathered_avg_reward_mean': '9.21215820', 'gathered_avg_reward_std': '1.53821480', 'gathered_video_score2_mq_reward_mean': '3.14428711', 'gathered_video_score2_mq_reward_std': '0.55506212', 'gathered_video_score2_overall_reward_mean': '9.20800781', 'gathered_video_score2_overall_reward_std': '1.53487515', 'gathered_video_score2_ta_reward_mean': '3.13281250', 'gathered_video_score2_ta_reward_std': '0.67052948', 'gathered_video_score2_vq_reward_mean': '2.93505859', 'gathered_video_score2_vq_reward_std': '0.66391009', 'gathered_ori_avg_reward_mean': '9.21215820', 'gathered_ori_avg_reward_std': '1.53821480'} +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 12:37:12.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_0/index_9828_seed_19026_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 12:38:33.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:33.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:39:47.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_0/index_9828_seed_19027_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 12:40:10.052 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:40:10.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 12:40:11.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.107 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 66 train_steps 2640 {'step': '66/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00401815', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '301.08825684', 'grpo_loop_time': '435.64062500', 'ref_model_time': '120.48604584', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84908956', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.98291016', 'token_credit_video_advantage_mean': '9.779e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.431e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00718889', 'kl_moving': '0.00718889', 'kl_stepwise': '-1.963e-08', 'gathered_avg_reward_mean': '8.98181152', 'gathered_avg_reward_std': '0.89241147', 'gathered_video_score2_mq_reward_mean': '3.07299805', 'gathered_video_score2_mq_reward_std': '0.22663076', 'gathered_video_score2_overall_reward_mean': '8.98291016', 'gathered_video_score2_overall_reward_std': '0.89255768', 'gathered_video_score2_ta_reward_mean': '3.12219238', 'gathered_video_score2_ta_reward_std': '0.55682969', 'gathered_video_score2_vq_reward_mean': '2.78662109', 'gathered_video_score2_vq_reward_std': '0.43306530', 'gathered_ori_avg_reward_mean': '8.98181152', 'gathered_ori_avg_reward_std': '0.89241147'} +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 12:51:30.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_0/index_8959_seed_28930_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 12:52:51.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:51.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:11.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_0/index_8959_seed_28931_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:54:31.863 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 12:54:32.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:58.894 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 67 train_steps 2680 {'step': '67/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00464977', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '305.14739990', 'grpo_loop_time': '439.27972412', 'ref_model_time': '121.56889343', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87292755', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.94287109', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.444e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00752715', 'kl_moving': '0.00752715', 'kl_stepwise': '-1.944e-08', 'gathered_avg_reward_mean': '8.94482422', 'gathered_avg_reward_std': '1.36484230', 'gathered_video_score2_mq_reward_mean': '3.12451172', 'gathered_video_score2_mq_reward_std': '0.42198321', 'gathered_video_score2_overall_reward_mean': '8.94287109', 'gathered_video_score2_overall_reward_std': '1.36246455', 'gathered_video_score2_ta_reward_mean': '3.06652832', 'gathered_video_score2_ta_reward_std': '0.64649177', 'gathered_video_score2_vq_reward_mean': '2.75378418', 'gathered_video_score2_vq_reward_std': '0.61234909', 'gathered_ori_avg_reward_mean': '8.94482422', 'gathered_ori_avg_reward_std': '1.36484230'} +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 13:05:59.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_0/index_4889_seed_4284_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 13:07:20.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:20.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:08:38.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_0/index_4889_seed_4285_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:09:03.456 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 13:09:04.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.125 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 68 train_steps 2720 {'step': '68/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00596165', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.36883545', 'grpo_loop_time': '435.01547241', 'ref_model_time': '120.63809967', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86891919', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03906250', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.453e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00792394', 'kl_moving': '0.00792394', 'kl_stepwise': '-2.013e-08', 'gathered_avg_reward_mean': '9.03784180', 'gathered_avg_reward_std': '1.21669281', 'gathered_video_score2_mq_reward_mean': '3.13281250', 'gathered_video_score2_mq_reward_std': '0.47114459', 'gathered_video_score2_overall_reward_mean': '9.03906250', 'gathered_video_score2_overall_reward_std': '1.21630609', 'gathered_video_score2_ta_reward_mean': '3.05224609', 'gathered_video_score2_ta_reward_std': '0.61982393', 'gathered_video_score2_vq_reward_mean': '2.85278320', 'gathered_video_score2_vq_reward_std': '0.55441618', 'gathered_ori_avg_reward_mean': '9.03784180', 'gathered_ori_avg_reward_std': '1.21669281'} +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 13:20:24.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:32.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_0/index_1455_seed_8802_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 13:21:51.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:51.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:00.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_0/index_1455_seed_8803_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:23:29.116 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 13:23:30.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.054 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 69 train_steps 2760 {'step': '69/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00681186', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.00573730', 'grpo_loop_time': '435.70715332', 'ref_model_time': '120.15560913', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88892299', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.67431641', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.427e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00751102', 'kl_moving': '0.00751102', 'kl_stepwise': '-1.899e-08', 'gathered_avg_reward_mean': '8.67614746', 'gathered_avg_reward_std': '1.63516223', 'gathered_video_score2_mq_reward_mean': '3.10217285', 'gathered_video_score2_mq_reward_std': '0.60644186', 'gathered_video_score2_overall_reward_mean': '8.67431641', 'gathered_video_score2_overall_reward_std': '1.63701987', 'gathered_video_score2_ta_reward_mean': '2.74121094', 'gathered_video_score2_ta_reward_std': '0.59759003', 'gathered_video_score2_vq_reward_mean': '2.83276367', 'gathered_video_score2_vq_reward_std': '0.72598964', 'gathered_ori_avg_reward_mean': '8.67614746', 'gathered_ori_avg_reward_std': '1.63516235'} +2026-05-30 13:34:48.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 13:34:48.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:56.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_0/index_3653_seed_9722_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 13:36:09.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:36:09.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:25.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_0/index_3653_seed_9723_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:37:47.913 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 13:37:48.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:06.861 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 70 train_steps 2800 {'step': '70/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00393851', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '304.85958862', 'grpo_loop_time': '432.54663086', 'ref_model_time': '122.96681976', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86502296', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.87597656', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.423e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00752172', 'kl_moving': '0.00752173', 'kl_stepwise': '-1.914e-08', 'gathered_avg_reward_mean': '8.87829590', 'gathered_avg_reward_std': '0.75983441', 'gathered_video_score2_mq_reward_mean': '3.12231445', 'gathered_video_score2_mq_reward_std': '0.27818912', 'gathered_video_score2_overall_reward_mean': '8.87597656', 'gathered_video_score2_overall_reward_std': '0.75932050', 'gathered_video_score2_ta_reward_mean': '2.93164062', 'gathered_video_score2_ta_reward_std': '0.61521733', 'gathered_video_score2_vq_reward_mean': '2.82434082', 'gathered_video_score2_vq_reward_std': '0.52909863', 'gathered_ori_avg_reward_mean': '8.87829590', 'gathered_ori_avg_reward_std': '0.75983441'} +2026-05-30 13:49:57.920 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 13:50:02.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_0/index_5473_seed_18408_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 13:51:31.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:31.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:52:48.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_0/index_5473_seed_18409_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 13:53:10.875 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:53:10.973 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 13:53:11.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.332 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 71 train_steps 2840 {'step': '71/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-10', 'grad_norm': '0.00557065', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '312.07366943', 'grpo_loop_time': '436.92779541', 'ref_model_time': '121.37475586', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.90730137', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.27392578', 'token_credit_video_advantage_mean': '8.848e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.391e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00747378', 'kl_moving': '0.00747379', 'kl_stepwise': '-1.876e-08', 'gathered_avg_reward_mean': '9.26940918', 'gathered_avg_reward_std': '1.26441646', 'gathered_video_score2_mq_reward_mean': '3.18542480', 'gathered_video_score2_mq_reward_std': '0.41633698', 'gathered_video_score2_overall_reward_mean': '9.27392578', 'gathered_video_score2_overall_reward_std': '1.26620042', 'gathered_video_score2_ta_reward_mean': '3.17346191', 'gathered_video_score2_ta_reward_std': '0.50160950', 'gathered_video_score2_vq_reward_mean': '2.91052246', 'gathered_video_score2_vq_reward_std': '0.61866128', 'gathered_ori_avg_reward_mean': '9.26940918', 'gathered_ori_avg_reward_std': '1.26441658'} +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 14:04:33.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:44.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_0/index_5902_seed_4454_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 14:06:01.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:06:01.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:19.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_0/index_5902_seed_4455_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:07:58.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 14:07:59.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.125 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 72 train_steps 2880 {'step': '72/10000', 'learning_rate': '0.00001000', 'loss': '-1.257e-08', 'grad_norm': '0.00498130', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '338.09494019', 'grpo_loop_time': '441.98263550', 'ref_model_time': '130.46066284', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86913520', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.99169922', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.446e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.257e-08', 'kl_loss': '0.00745321', 'kl_moving': '0.00745321', 'kl_stepwise': '-1.967e-08', 'gathered_avg_reward_mean': '8.99169922', 'gathered_avg_reward_std': '1.17242658', 'gathered_video_score2_mq_reward_mean': '3.05029297', 'gathered_video_score2_mq_reward_std': '0.44091526', 'gathered_video_score2_overall_reward_mean': '8.99169922', 'gathered_video_score2_overall_reward_std': '1.17329657', 'gathered_video_score2_ta_reward_mean': '3.13452148', 'gathered_video_score2_ta_reward_std': '0.61896265', 'gathered_video_score2_vq_reward_mean': '2.80688477', 'gathered_video_score2_vq_reward_std': '0.54264456', 'gathered_ori_avg_reward_mean': '8.99169922', 'gathered_ori_avg_reward_std': '1.17242658'} +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 14:19:43.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_0/index_9363_seed_11532_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 14:21:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:21:16.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:22:39.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_0/index_9363_seed_11533_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:23:12.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 14:23:13.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:37.955 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 73 train_steps 2920 {'step': '73/10000', 'learning_rate': '0.00001000', 'loss': '1.676e-08', 'grad_norm': '0.00531907', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '333.01974487', 'grpo_loop_time': '438.41967773', 'ref_model_time': '121.70159149', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85791552', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.12988281', 'token_credit_video_advantage_mean': '-1.187e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.456e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.676e-08', 'kl_loss': '0.00724727', 'kl_moving': '0.00724728', 'kl_stepwise': '-1.929e-08', 'gathered_avg_reward_mean': '9.12377930', 'gathered_avg_reward_std': '1.13538074', 'gathered_video_score2_mq_reward_mean': '3.14526367', 'gathered_video_score2_mq_reward_std': '0.39363760', 'gathered_video_score2_overall_reward_mean': '9.12988281', 'gathered_video_score2_overall_reward_std': '1.13580179', 'gathered_video_score2_ta_reward_mean': '3.06066895', 'gathered_video_score2_ta_reward_std': '0.60277998', 'gathered_video_score2_vq_reward_mean': '2.91784668', 'gathered_video_score2_vq_reward_std': '0.59187102', 'gathered_ori_avg_reward_mean': '9.12377930', 'gathered_ori_avg_reward_std': '1.13538086'} +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 14:34:38.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_0/index_4474_seed_30408_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 14:36:02.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:36:02.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:16.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_0/index_4474_seed_30409_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:37:53.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 14:37:54.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.695 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 74 train_steps 2960 {'step': '74/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00395323', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '324.13638306', 'grpo_loop_time': '433.47921753', 'ref_model_time': '126.71461487', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88783556', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.97363281', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.523e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00676167', 'kl_moving': '0.00676167', 'kl_stepwise': '-1.970e-08', 'gathered_avg_reward_mean': '8.97290039', 'gathered_avg_reward_std': '1.45057964', 'gathered_video_score2_mq_reward_mean': '3.09741211', 'gathered_video_score2_mq_reward_std': '0.48531711', 'gathered_video_score2_overall_reward_mean': '8.97363281', 'gathered_video_score2_overall_reward_std': '1.45016313', 'gathered_video_score2_ta_reward_mean': '2.94946289', 'gathered_video_score2_ta_reward_std': '0.64965993', 'gathered_video_score2_vq_reward_mean': '2.92602539', 'gathered_video_score2_vq_reward_std': '0.52614516', 'gathered_ori_avg_reward_mean': '8.97290039', 'gathered_ori_avg_reward_std': '1.45057964'} +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 14:49:19.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:28.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_0/index_2055_seed_18410_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 14:50:43.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:43.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:00.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_0/index_2055_seed_18411_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:52:25.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 14:52:26.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:03:50.458 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 75 train_steps 3000 {'step': '75/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00349371', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '311.10656738', 'grpo_loop_time': '437.57446289', 'ref_model_time': '122.75245667', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87664145', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.30761719', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.395e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.00690167', 'kl_moving': '0.00690167', 'kl_stepwise': '-1.870e-08', 'gathered_avg_reward_mean': '9.30432129', 'gathered_avg_reward_std': '1.28977883', 'gathered_video_score2_mq_reward_mean': '3.17944336', 'gathered_video_score2_mq_reward_std': '0.39623889', 'gathered_video_score2_overall_reward_mean': '9.30761719', 'gathered_video_score2_overall_reward_std': '1.29405069', 'gathered_video_score2_ta_reward_mean': '3.07458496', 'gathered_video_score2_ta_reward_std': '0.66933215', 'gathered_video_score2_vq_reward_mean': '3.05029297', 'gathered_video_score2_vq_reward_std': '0.49855763', 'gathered_ori_avg_reward_mean': '9.30432129', 'gathered_ori_avg_reward_std': '1.28977883'} +2026-05-30 15:04:58.418 | INFO | __main__:save_checkpoint:1917 - Timer checkpoint saved at step 75 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-75 +2026-05-30 15:04:58.562 | INFO | __main__:_write_timer_marker:1761 - Timer resume marker updated at step 75: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-75 +2026-05-30 15:05:01.245 | INFO | __main__:_prune_old_timer_checkpoints:1776 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-50 +2026-05-30 15:05:01.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 15:05:01.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:10.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_0/index_2964_seed_16199_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 15:06:24.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:24.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:07:40.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_0/index_2964_seed_16200_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:08:03.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 15:08:04.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.640 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 76 train_steps 3040 {'step': '76/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00458203', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '311.11111450', 'grpo_loop_time': '436.20605469', 'ref_model_time': '126.71891785', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89623201', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.78222656', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.430e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00760994', 'kl_moving': '0.00760994', 'kl_stepwise': '-2.008e-08', 'gathered_avg_reward_mean': '8.78222656', 'gathered_avg_reward_std': '1.12177205', 'gathered_video_score2_mq_reward_mean': '3.03796387', 'gathered_video_score2_mq_reward_std': '0.33582640', 'gathered_video_score2_overall_reward_mean': '8.78222656', 'gathered_video_score2_overall_reward_std': '1.12236524', 'gathered_video_score2_ta_reward_mean': '3.02661133', 'gathered_video_score2_ta_reward_std': '0.67965204', 'gathered_video_score2_vq_reward_mean': '2.71765137', 'gathered_video_score2_vq_reward_std': '0.53348362', 'gathered_ori_avg_reward_mean': '8.78222656', 'gathered_ori_avg_reward_std': '1.12177193'} +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 15:19:35.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_0/index_6685_seed_4210_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 15:20:56.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:56.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:14.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_0/index_6685_seed_4211_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:22:37.436 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 15:22:38.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.658 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 77 train_steps 3080 {'step': '77/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00451039', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '312.50656128', 'grpo_loop_time': '435.34136963', 'ref_model_time': '127.52886200', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84229702', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.39892578', 'token_credit_video_advantage_mean': '-8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.496e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00701217', 'kl_moving': '0.00701218', 'kl_stepwise': '-2.013e-08', 'gathered_avg_reward_mean': '9.39782715', 'gathered_avg_reward_std': '1.28862095', 'gathered_video_score2_mq_reward_mean': '3.27807617', 'gathered_video_score2_mq_reward_std': '0.41650754', 'gathered_video_score2_overall_reward_mean': '9.39892578', 'gathered_video_score2_overall_reward_std': '1.28678477', 'gathered_video_score2_ta_reward_mean': '2.97827148', 'gathered_video_score2_ta_reward_std': '0.72124606', 'gathered_video_score2_vq_reward_mean': '3.14147949', 'gathered_video_score2_vq_reward_std': '0.54402179', 'gathered_ori_avg_reward_mean': '9.39782715', 'gathered_ori_avg_reward_std': '1.28862095'} +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 15:34:11.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_0/index_8076_seed_31063_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 15:35:36.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:36.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:36:53.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_0/index_8076_seed_31064_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:37:18.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 15:37:19.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:47.937 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 78 train_steps 3120 {'step': '78/10000', 'learning_rate': '0.00001000', 'loss': '1.909e-08', 'grad_norm': '0.00599822', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '313.78207397', 'grpo_loop_time': '437.14013672', 'ref_model_time': '124.31060028', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.82951039', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.00683594', 'token_credit_video_advantage_mean': '-4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.909e-08', 'kl_loss': '0.00773323', 'kl_moving': '0.00773324', 'kl_stepwise': '-1.907e-08', 'gathered_avg_reward_mean': '9.00280762', 'gathered_avg_reward_std': '0.83731943', 'gathered_video_score2_mq_reward_mean': '3.17504883', 'gathered_video_score2_mq_reward_std': '0.31930074', 'gathered_video_score2_overall_reward_mean': '9.00683594', 'gathered_video_score2_overall_reward_std': '0.83875233', 'gathered_video_score2_ta_reward_mean': '2.97131348', 'gathered_video_score2_ta_reward_std': '0.53867149', 'gathered_video_score2_vq_reward_mean': '2.85644531', 'gathered_video_score2_vq_reward_std': '0.55853140', 'gathered_ori_avg_reward_mean': '9.00280762', 'gathered_ori_avg_reward_std': '0.83731949'} +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 15:48:48.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:57.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_0/index_2217_seed_24417_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 15:50:14.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:50:14.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:25.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_0/index_2217_seed_24418_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:51:50.298 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 15:51:51.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.631 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 79 train_steps 3160 {'step': '79/10000', 'learning_rate': '0.00001000', 'loss': '-5.122e-09', 'grad_norm': '0.01584567', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.77825928', 'grpo_loop_time': '436.45898438', 'ref_model_time': '123.37480164', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86132652', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.74072266', 'token_credit_video_advantage_mean': '-4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.469e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.122e-09', 'kl_loss': '0.00793886', 'kl_moving': '0.00793886', 'kl_stepwise': '-1.968e-08', 'gathered_avg_reward_mean': '8.73388672', 'gathered_avg_reward_std': '1.07100546', 'gathered_video_score2_mq_reward_mean': '3.03186035', 'gathered_video_score2_mq_reward_std': '0.38090020', 'gathered_video_score2_overall_reward_mean': '8.74072266', 'gathered_video_score2_overall_reward_std': '1.07581103', 'gathered_video_score2_ta_reward_mean': '2.90197754', 'gathered_video_score2_ta_reward_std': '0.56663340', 'gathered_video_score2_vq_reward_mean': '2.80004883', 'gathered_video_score2_vq_reward_std': '0.54950488', 'gathered_ori_avg_reward_mean': '8.73388672', 'gathered_ori_avg_reward_std': '1.07100534'} +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 16:03:16.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_0/index_2237_seed_22609_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 16:04:40.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:40.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:05:57.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_0/index_2237_seed_22610_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:06:19.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 16:06:20.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:17:52.361 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 80 train_steps 3200 {'step': '80/10000', 'learning_rate': '0.00001000', 'loss': '1.024e-08', 'grad_norm': '0.00603401', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '310.19454956', 'grpo_loop_time': '437.94869995', 'ref_model_time': '125.06247711', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88641250', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.61767578', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.410e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.024e-08', 'kl_loss': '0.00754499', 'kl_moving': '0.00754499', 'kl_stepwise': '-2.001e-08', 'gathered_avg_reward_mean': '8.61242676', 'gathered_avg_reward_std': '0.96937251', 'gathered_video_score2_mq_reward_mean': '3.00659180', 'gathered_video_score2_mq_reward_std': '0.31759524', 'gathered_video_score2_overall_reward_mean': '8.61767578', 'gathered_video_score2_overall_reward_std': '0.97281688', 'gathered_video_score2_ta_reward_mean': '2.77099609', 'gathered_video_score2_ta_reward_std': '0.55480587', 'gathered_video_score2_vq_reward_mean': '2.83483887', 'gathered_video_score2_vq_reward_std': '0.53361768', 'gathered_ori_avg_reward_mean': '8.61242676', 'gathered_ori_avg_reward_std': '0.96937257'} +2026-05-30 16:18:43.847 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 16:18:48.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_0/index_357_seed_1772_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 16:20:19.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:19.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:21:33.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_0/index_357_seed_1773_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 16:22:04.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:22:04.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 16:22:05.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.423 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 81 train_steps 3240 {'step': '81/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00508915', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '323.12512207', 'grpo_loop_time': '437.36328125', 'ref_model_time': '124.36315155', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86666960', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03857422', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.432e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00743154', 'kl_moving': '0.00743154', 'kl_stepwise': '-1.943e-08', 'gathered_avg_reward_mean': '9.03881836', 'gathered_avg_reward_std': '0.87814093', 'gathered_video_score2_mq_reward_mean': '3.17529297', 'gathered_video_score2_mq_reward_std': '0.29546422', 'gathered_video_score2_overall_reward_mean': '9.03857422', 'gathered_video_score2_overall_reward_std': '0.87492573', 'gathered_video_score2_ta_reward_mean': '3.09716797', 'gathered_video_score2_ta_reward_std': '0.50110102', 'gathered_video_score2_vq_reward_mean': '2.76635742', 'gathered_video_score2_vq_reward_std': '0.54630399', 'gathered_ori_avg_reward_mean': '9.03881836', 'gathered_ori_avg_reward_std': '0.87814093'} +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 16:33:33.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:33.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:43.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_0/index_6482_seed_17570_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 16:34:59.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:59.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:16.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_0/index_6482_seed_17571_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:36:39.642 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 16:36:40.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.247 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 82 train_steps 3280 {'step': '82/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00422213', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '312.74148560', 'grpo_loop_time': '435.75677490', 'ref_model_time': '124.43354034', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85764295', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.04248047', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.436e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00699377', 'kl_moving': '0.00699377', 'kl_stepwise': '-1.916e-08', 'gathered_avg_reward_mean': '9.04736328', 'gathered_avg_reward_std': '1.03169739', 'gathered_video_score2_mq_reward_mean': '3.14184570', 'gathered_video_score2_mq_reward_std': '0.37398449', 'gathered_video_score2_overall_reward_mean': '9.04248047', 'gathered_video_score2_overall_reward_std': '1.02875185', 'gathered_video_score2_ta_reward_mean': '2.99792480', 'gathered_video_score2_ta_reward_std': '0.65540838', 'gathered_video_score2_vq_reward_mean': '2.90759277', 'gathered_video_score2_vq_reward_std': '0.57287449', 'gathered_ori_avg_reward_mean': '9.04736328', 'gathered_ori_avg_reward_std': '1.03169739'} +2026-05-30 16:48:15.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 16:48:15.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:24.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_0/index_4776_seed_6618_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 16:49:38.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:38.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:50:53.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_0/index_4776_seed_6619_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:51:15.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 16:51:16.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:54.660 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 83 train_steps 3320 {'step': '83/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00522345', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '306.52819824', 'grpo_loop_time': '443.00076294', 'ref_model_time': '124.03240967', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85313594', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.02343750', 'token_credit_video_advantage_mean': '-7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.401e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00689306', 'kl_moving': '0.00689306', 'kl_stepwise': '-1.939e-08', 'gathered_avg_reward_mean': '9.01989746', 'gathered_avg_reward_std': '1.15723276', 'gathered_video_score2_mq_reward_mean': '3.19970703', 'gathered_video_score2_mq_reward_std': '0.37449580', 'gathered_video_score2_overall_reward_mean': '9.02343750', 'gathered_video_score2_overall_reward_std': '1.15775347', 'gathered_video_score2_ta_reward_mean': '2.95764160', 'gathered_video_score2_ta_reward_std': '0.63200390', 'gathered_video_score2_vq_reward_mean': '2.86254883', 'gathered_video_score2_vq_reward_std': '0.53953236', 'gathered_ori_avg_reward_mean': '9.01989746', 'gathered_ori_avg_reward_std': '1.15723264'} +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 17:02:55.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:04.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_0/index_306_seed_30904_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 17:04:23.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:23.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:33.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_0/index_306_seed_30905_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:05:56.377 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 17:05:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:34.573 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 84 train_steps 3360 {'step': '84/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00435070', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '311.49725342', 'grpo_loop_time': '439.59918213', 'ref_model_time': '127.91097260', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86507589', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01757812', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.471e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00689303', 'kl_moving': '0.00689303', 'kl_stepwise': '-1.869e-08', 'gathered_avg_reward_mean': '9.02197266', 'gathered_avg_reward_std': '1.26533175', 'gathered_video_score2_mq_reward_mean': '3.13537598', 'gathered_video_score2_mq_reward_std': '0.45173812', 'gathered_video_score2_overall_reward_mean': '9.01757812', 'gathered_video_score2_overall_reward_std': '1.26290786', 'gathered_video_score2_ta_reward_mean': '3.00744629', 'gathered_video_score2_ta_reward_std': '0.66392523', 'gathered_video_score2_vq_reward_mean': '2.87915039', 'gathered_video_score2_vq_reward_std': '0.58243251', 'gathered_ori_avg_reward_mean': '9.02197266', 'gathered_ori_avg_reward_std': '1.26533175'} +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 17:17:35.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_0/index_7318_seed_27899_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 17:19:00.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:19:01.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:15.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_0/index_7318_seed_27900_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:20:37.463 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 17:20:38.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.015 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 85 train_steps 3400 {'step': '85/10000', 'learning_rate': '0.00001000', 'loss': '1.490e-08', 'grad_norm': '0.00623025', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.66049194', 'grpo_loop_time': '437.95285034', 'ref_model_time': '122.94744110', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86289984', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.59863281', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.492e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.490e-08', 'kl_loss': '0.00721119', 'kl_moving': '0.00721119', 'kl_stepwise': '-1.997e-08', 'gathered_avg_reward_mean': '8.59777832', 'gathered_avg_reward_std': '1.13732600', 'gathered_video_score2_mq_reward_mean': '2.96960449', 'gathered_video_score2_mq_reward_std': '0.40199780', 'gathered_video_score2_overall_reward_mean': '8.59863281', 'gathered_video_score2_overall_reward_std': '1.13541210', 'gathered_video_score2_ta_reward_mean': '2.97717285', 'gathered_video_score2_ta_reward_std': '0.67067069', 'gathered_video_score2_vq_reward_mean': '2.65100098', 'gathered_video_score2_vq_reward_std': '0.51283586', 'gathered_ori_avg_reward_mean': '8.59777832', 'gathered_ori_avg_reward_std': '1.13732600'} +2026-05-30 17:32:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 17:32:10.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_0/index_7637_seed_15724_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 17:33:39.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:39.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:00.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_0/index_7637_seed_15725_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:35:25.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 17:35:26.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.227 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 86 train_steps 3440 {'step': '86/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00425077', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '321.44390869', 'grpo_loop_time': '438.77767944', 'ref_model_time': '123.82201385', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86214143', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.39062500', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.483e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00755565', 'kl_moving': '0.00755565', 'kl_stepwise': '-2.067e-08', 'gathered_avg_reward_mean': '9.39270020', 'gathered_avg_reward_std': '1.04686880', 'gathered_video_score2_mq_reward_mean': '3.24145508', 'gathered_video_score2_mq_reward_std': '0.29180667', 'gathered_video_score2_overall_reward_mean': '9.39062500', 'gathered_video_score2_overall_reward_std': '1.04370725', 'gathered_video_score2_ta_reward_mean': '3.26013184', 'gathered_video_score2_ta_reward_std': '0.56993729', 'gathered_video_score2_vq_reward_mean': '2.89111328', 'gathered_video_score2_vq_reward_std': '0.49702308', 'gathered_ori_avg_reward_mean': '9.39270020', 'gathered_ori_avg_reward_std': '1.04686880'} +2026-05-30 17:46:54.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 17:46:54.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_0/index_7339_seed_10900_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 17:48:20.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:20.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:49:38.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_0/index_7339_seed_10901_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:50:01.520 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 17:50:02.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:34.713 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 87 train_steps 3480 {'step': '87/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00539804', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '313.46365356', 'grpo_loop_time': '440.13830566', 'ref_model_time': '124.22696686', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87416279', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.77246094', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.417e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00841283', 'kl_moving': '0.00841283', 'kl_stepwise': '-1.898e-08', 'gathered_avg_reward_mean': '8.77001953', 'gathered_avg_reward_std': '1.19083524', 'gathered_video_score2_mq_reward_mean': '3.09423828', 'gathered_video_score2_mq_reward_std': '0.35388738', 'gathered_video_score2_overall_reward_mean': '8.77246094', 'gathered_video_score2_overall_reward_std': '1.19114971', 'gathered_video_score2_ta_reward_mean': '2.86633301', 'gathered_video_score2_ta_reward_std': '0.67138106', 'gathered_video_score2_vq_reward_mean': '2.80944824', 'gathered_video_score2_vq_reward_std': '0.54201704', 'gathered_ori_avg_reward_mean': '8.77001953', 'gathered_ori_avg_reward_std': '1.19083524'} +2026-05-30 18:01:35.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 18:01:35.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:44.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_0/index_1372_seed_1463_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 18:02:59.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:59.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:17.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_0/index_1372_seed_1464_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:04:39.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 18:04:40.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.443 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 88 train_steps 3520 {'step': '88/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00657530', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '312.18322754', 'grpo_loop_time': '438.95632935', 'ref_model_time': '124.99347687', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85452855', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.28222656', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.461e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.211e-08', 'kl_loss': '0.00824794', 'kl_moving': '0.00824795', 'kl_stepwise': '-1.945e-08', 'gathered_avg_reward_mean': '8.28161621', 'gathered_avg_reward_std': '1.18030608', 'gathered_video_score2_mq_reward_mean': '2.89880371', 'gathered_video_score2_mq_reward_std': '0.39588845', 'gathered_video_score2_overall_reward_mean': '8.28222656', 'gathered_video_score2_overall_reward_std': '1.18274868', 'gathered_video_score2_ta_reward_mean': '2.84033203', 'gathered_video_score2_ta_reward_std': '0.57311302', 'gathered_video_score2_vq_reward_mean': '2.54248047', 'gathered_video_score2_vq_reward_std': '0.50210297', 'gathered_ori_avg_reward_mean': '8.28161621', 'gathered_ori_avg_reward_std': '1.18030608'} +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 18:16:17.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:17.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_0/index_8870_seed_20304_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 18:17:41.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:41.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:00.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_0/index_8870_seed_20305_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 18:19:30.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:19:31.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 18:19:32.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.198 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 89 train_steps 3560 {'step': '89/10000', 'learning_rate': '0.00001000', 'loss': '-3.260e-09', 'grad_norm': '0.00404601', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '322.92083740', 'grpo_loop_time': '441.70959473', 'ref_model_time': '126.99559784', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87049901', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.96923828', 'token_credit_video_advantage_mean': '-6.985e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.409e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.260e-09', 'kl_loss': '0.00767261', 'kl_moving': '0.00767261', 'kl_stepwise': '-1.938e-08', 'gathered_avg_reward_mean': '8.97009277', 'gathered_avg_reward_std': '1.02774990', 'gathered_video_score2_mq_reward_mean': '3.18359375', 'gathered_video_score2_mq_reward_std': '0.33175540', 'gathered_video_score2_overall_reward_mean': '8.96923828', 'gathered_video_score2_overall_reward_std': '1.02971733', 'gathered_video_score2_ta_reward_mean': '2.96362305', 'gathered_video_score2_ta_reward_std': '0.59940290', 'gathered_video_score2_vq_reward_mean': '2.82287598', 'gathered_video_score2_vq_reward_std': '0.47427177', 'gathered_ori_avg_reward_mean': '8.97009277', 'gathered_ori_avg_reward_std': '1.02774990'} +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 18:31:18.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:30.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_0/index_3239_seed_18003_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 18:32:46.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:46.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:14.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_0/index_3239_seed_18004_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 18:34:45.308 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:34:45.471 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 18:34:46.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:46:22.794 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 90 train_steps 3600 {'step': '90/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00425579', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '334.58862305', 'grpo_loop_time': '440.86798096', 'ref_model_time': '125.24485016', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87546384', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.04443359', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.442e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00705699', 'kl_moving': '0.00705700', 'kl_stepwise': '-1.951e-08', 'gathered_avg_reward_mean': '9.04663086', 'gathered_avg_reward_std': '1.30403388', 'gathered_video_score2_mq_reward_mean': '3.17504883', 'gathered_video_score2_mq_reward_std': '0.48231652', 'gathered_video_score2_overall_reward_mean': '9.04443359', 'gathered_video_score2_overall_reward_std': '1.30046809', 'gathered_video_score2_ta_reward_mean': '2.87890625', 'gathered_video_score2_ta_reward_std': '0.72335136', 'gathered_video_score2_vq_reward_mean': '2.99267578', 'gathered_video_score2_vq_reward_std': '0.57623291', 'gathered_ori_avg_reward_mean': '9.04663086', 'gathered_ori_avg_reward_std': '1.30403388'} +2026-05-30 18:47:18.026 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 18:47:24.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:35.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_0/index_9416_seed_10361_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 18:48:50.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:50.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:12.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_0/index_9416_seed_10362_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 18:50:38.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:50:38.709 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 18:50:39.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.138 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 91 train_steps 3640 {'step': '91/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00418312', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '321.50341797', 'grpo_loop_time': '440.00659180', 'ref_model_time': '124.28290558', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87132257', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24462891', 'token_credit_video_advantage_mean': '7.916e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.453e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00655295', 'kl_moving': '0.00655295', 'kl_stepwise': '-1.897e-08', 'gathered_avg_reward_mean': '9.24877930', 'gathered_avg_reward_std': '0.96896780', 'gathered_video_score2_mq_reward_mean': '3.17968750', 'gathered_video_score2_mq_reward_std': '0.30955645', 'gathered_video_score2_overall_reward_mean': '9.24462891', 'gathered_video_score2_overall_reward_std': '0.97045833', 'gathered_video_score2_ta_reward_mean': '3.09912109', 'gathered_video_score2_ta_reward_std': '0.60887533', 'gathered_video_score2_vq_reward_mean': '2.96997070', 'gathered_video_score2_vq_reward_std': '0.42068249', 'gathered_ori_avg_reward_mean': '9.24877930', 'gathered_ori_avg_reward_std': '0.96896774'} +2026-05-30 19:02:13.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 19:02:13.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_0/index_827_seed_17315_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 19:03:43.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:43.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:01.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_0/index_827_seed_17316_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:05:34.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 19:05:34.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.305 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 92 train_steps 3680 {'step': '92/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00427702', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '330.33105469', 'grpo_loop_time': '439.52883911', 'ref_model_time': '127.26576996', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87588137', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.89208984', 'token_credit_video_advantage_mean': '8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00692372', 'kl_moving': '0.00692372', 'kl_stepwise': '-1.934e-08', 'gathered_avg_reward_mean': '8.89062500', 'gathered_avg_reward_std': '1.00102758', 'gathered_video_score2_mq_reward_mean': '3.14111328', 'gathered_video_score2_mq_reward_std': '0.31345034', 'gathered_video_score2_overall_reward_mean': '8.89208984', 'gathered_video_score2_overall_reward_std': '1.00260484', 'gathered_video_score2_ta_reward_mean': '2.93530273', 'gathered_video_score2_ta_reward_std': '0.60845482', 'gathered_video_score2_vq_reward_mean': '2.81420898', 'gathered_video_score2_vq_reward_std': '0.47984403', 'gathered_ori_avg_reward_mean': '8.89062500', 'gathered_ori_avg_reward_std': '1.00102758'} +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 19:17:12.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_0/index_6760_seed_17098_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 19:18:35.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:35.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:19:58.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_0/index_6760_seed_17099_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:20:25.855 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 19:20:26.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:54.911 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 93 train_steps 3720 {'step': '93/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00417444', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '318.77810669', 'grpo_loop_time': '439.42361450', 'ref_model_time': '123.31369781', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84829986', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.10644531', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.406e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.00679202', 'kl_moving': '0.00679203', 'kl_stepwise': '-1.917e-08', 'gathered_avg_reward_mean': '9.10778809', 'gathered_avg_reward_std': '0.87333763', 'gathered_video_score2_mq_reward_mean': '3.19726562', 'gathered_video_score2_mq_reward_std': '0.31599003', 'gathered_video_score2_overall_reward_mean': '9.10644531', 'gathered_video_score2_overall_reward_std': '0.87391323', 'gathered_video_score2_ta_reward_mean': '2.95312500', 'gathered_video_score2_ta_reward_std': '0.60732609', 'gathered_video_score2_vq_reward_mean': '2.95739746', 'gathered_video_score2_vq_reward_std': '0.48221809', 'gathered_ori_avg_reward_mean': '9.10778809', 'gathered_ori_avg_reward_std': '0.87333763'} +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 19:31:55.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:06.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_0/index_4904_seed_17667_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 19:33:20.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:20.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:34:55.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_0/index_4904_seed_17668_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 19:35:23.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:35:23.161 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 19:35:24.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:02.829 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 94 train_steps 3760 {'step': '94/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00317987', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '337.79650879', 'grpo_loop_time': '439.30734253', 'ref_model_time': '127.41754913', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84288108', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.08154297', 'token_credit_video_advantage_mean': '-3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.429e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00695214', 'kl_moving': '0.00695214', 'kl_stepwise': '-1.933e-08', 'gathered_avg_reward_mean': '9.08032227', 'gathered_avg_reward_std': '0.76051271', 'gathered_video_score2_mq_reward_mean': '3.15478516', 'gathered_video_score2_mq_reward_std': '0.30901030', 'gathered_video_score2_overall_reward_mean': '9.08154297', 'gathered_video_score2_overall_reward_std': '0.75826776', 'gathered_video_score2_ta_reward_mean': '3.05957031', 'gathered_video_score2_ta_reward_std': '0.59879053', 'gathered_video_score2_vq_reward_mean': '2.86596680', 'gathered_video_score2_vq_reward_std': '0.46412763', 'gathered_ori_avg_reward_mean': '9.08032227', 'gathered_ori_avg_reward_std': '0.76051271'} +2026-05-30 19:47:03.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 19:47:03.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_0/index_2342_seed_31533_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 19:48:28.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:28.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:49:47.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_0/index_2342_seed_31534_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 19:50:10.584 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:50:10.705 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 19:50:11.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:45.951 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 95 train_steps 3800 {'step': '95/10000', 'learning_rate': '0.00001000', 'loss': '-1.048e-08', 'grad_norm': '0.00447842', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '313.29757690', 'grpo_loop_time': '442.85891724', 'ref_model_time': '123.61854553', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89438385', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.51049805', 'token_credit_video_advantage_mean': '-3.958e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.048e-08', 'kl_loss': '0.00710088', 'kl_moving': '0.00710088', 'kl_stepwise': '-1.926e-08', 'gathered_avg_reward_mean': '8.50708008', 'gathered_avg_reward_std': '1.42185533', 'gathered_video_score2_mq_reward_mean': '2.93701172', 'gathered_video_score2_mq_reward_std': '0.39947942', 'gathered_video_score2_overall_reward_mean': '8.51049805', 'gathered_video_score2_overall_reward_std': '1.42017853', 'gathered_video_score2_ta_reward_mean': '2.88073730', 'gathered_video_score2_ta_reward_std': '0.73875159', 'gathered_video_score2_vq_reward_mean': '2.68933105', 'gathered_video_score2_vq_reward_std': '0.57372153', 'gathered_ori_avg_reward_mean': '8.50708008', 'gathered_ori_avg_reward_std': '1.42185533'} +2026-05-30 20:01:46.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 20:01:46.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:57.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_0/index_6549_seed_31222_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 20:03:14.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:03:14.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:29.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_0/index_6549_seed_31223_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:04:55.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 20:04:56.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.019 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 96 train_steps 3840 {'step': '96/10000', 'learning_rate': '0.00001000', 'loss': '1.397e-08', 'grad_norm': '0.00640652', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '318.60351562', 'grpo_loop_time': '438.79730225', 'ref_model_time': '127.05370331', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87770545', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.65283203', 'token_credit_video_advantage_mean': '-3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.381e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.350e-08', 'kl_loss': '0.00750481', 'kl_moving': '0.00750481', 'kl_stepwise': '-1.854e-08', 'gathered_avg_reward_mean': '8.65209961', 'gathered_avg_reward_std': '1.13745761', 'gathered_video_score2_mq_reward_mean': '3.08227539', 'gathered_video_score2_mq_reward_std': '0.34864503', 'gathered_video_score2_overall_reward_mean': '8.65283203', 'gathered_video_score2_overall_reward_std': '1.13308346', 'gathered_video_score2_ta_reward_mean': '2.77343750', 'gathered_video_score2_ta_reward_std': '0.62764031', 'gathered_video_score2_vq_reward_mean': '2.79638672', 'gathered_video_score2_vq_reward_std': '0.61211675', 'gathered_ori_avg_reward_mean': '8.65209961', 'gathered_ori_avg_reward_std': '1.13745761'} +2026-05-30 20:16:29.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 20:16:29.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:41.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_0/index_8016_seed_1341_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 20:17:56.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:56.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:19.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_0/index_8016_seed_1342_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:19:46.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 20:19:47.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.495 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 97 train_steps 3880 {'step': '97/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00419811', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '323.57693481', 'grpo_loop_time': '439.43362427', 'ref_model_time': '123.85105133', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85858822', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.51220703', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.496e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00746832', 'kl_moving': '0.00746832', 'kl_stepwise': '-1.985e-08', 'gathered_avg_reward_mean': '9.51074219', 'gathered_avg_reward_std': '1.09725094', 'gathered_video_score2_mq_reward_mean': '3.31982422', 'gathered_video_score2_mq_reward_std': '0.39701769', 'gathered_video_score2_overall_reward_mean': '9.51220703', 'gathered_video_score2_overall_reward_std': '1.09215307', 'gathered_video_score2_ta_reward_mean': '3.28198242', 'gathered_video_score2_ta_reward_std': '0.55249918', 'gathered_video_score2_vq_reward_mean': '2.90893555', 'gathered_video_score2_vq_reward_std': '0.55148417', 'gathered_ori_avg_reward_mean': '9.51074219', 'gathered_ori_avg_reward_std': '1.09725082'} +2026-05-30 20:31:19.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 20:31:19.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:19.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_0/index_5950_seed_24152_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 20:32:42.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:42.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:33:59.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_0/index_5950_seed_24153_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:34:23.225 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 20:34:24.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:02.826 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 98 train_steps 3920 {'step': '98/10000', 'learning_rate': '0.00001000', 'loss': '1.024e-08', 'grad_norm': '0.00327620', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '314.12927246', 'grpo_loop_time': '441.77392578', 'ref_model_time': '128.10758972', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86718619', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.11376953', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.382e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.024e-08', 'kl_loss': '0.00701776', 'kl_moving': '0.00701776', 'kl_stepwise': '-1.887e-08', 'gathered_avg_reward_mean': '9.11682129', 'gathered_avg_reward_std': '1.23930538', 'gathered_video_score2_mq_reward_mean': '3.18823242', 'gathered_video_score2_mq_reward_std': '0.40727624', 'gathered_video_score2_overall_reward_mean': '9.11376953', 'gathered_video_score2_overall_reward_std': '1.24337065', 'gathered_video_score2_ta_reward_mean': '2.84765625', 'gathered_video_score2_ta_reward_std': '0.73106384', 'gathered_video_score2_vq_reward_mean': '3.08093262', 'gathered_video_score2_vq_reward_std': '0.45835099', 'gathered_ori_avg_reward_mean': '9.11682129', 'gathered_ori_avg_reward_std': '1.23930550'} +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 20:46:03.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_0/index_6234_seed_10501_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 20:47:23.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:23.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:48:49.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_0/index_6234_seed_10502_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:49:14.625 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 20:49:15.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.543 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 99 train_steps 3960 {'step': '99/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00406593', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '316.44723511', 'grpo_loop_time': '441.06033325', 'ref_model_time': '122.73826599', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86371100', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.10888672', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.366e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00745535', 'kl_moving': '0.00745536', 'kl_stepwise': '-1.942e-08', 'gathered_avg_reward_mean': '9.10681152', 'gathered_avg_reward_std': '0.99102390', 'gathered_video_score2_mq_reward_mean': '3.11950684', 'gathered_video_score2_mq_reward_std': '0.34168157', 'gathered_video_score2_overall_reward_mean': '9.10888672', 'gathered_video_score2_overall_reward_std': '0.98889083', 'gathered_video_score2_ta_reward_mean': '3.01354980', 'gathered_video_score2_ta_reward_std': '0.54077625', 'gathered_video_score2_vq_reward_mean': '2.97375488', 'gathered_video_score2_vq_reward_std': '0.53236252', 'gathered_ori_avg_reward_mean': '9.10681152', 'gathered_ori_avg_reward_std': '0.99102384'} +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 21:00:47.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:47.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:56.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_0/index_9805_seed_298_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 21:02:08.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:02:08.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:31.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_0/index_9805_seed_299_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:03:57.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 21:03:58.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:15:25.854 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 100 train_steps 4000 {'step': '100/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00415137', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '317.02593994', 'grpo_loop_time': '437.87207031', 'ref_model_time': '125.02745819', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.82568759', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91650391', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.392e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00772091', 'kl_moving': '0.00772091', 'kl_stepwise': '-1.930e-08', 'gathered_avg_reward_mean': '8.91845703', 'gathered_avg_reward_std': '1.01360595', 'gathered_video_score2_mq_reward_mean': '3.13940430', 'gathered_video_score2_mq_reward_std': '0.31368557', 'gathered_video_score2_overall_reward_mean': '8.91650391', 'gathered_video_score2_overall_reward_std': '1.01446831', 'gathered_video_score2_ta_reward_mean': '2.94482422', 'gathered_video_score2_ta_reward_std': '0.56780684', 'gathered_video_score2_vq_reward_mean': '2.83422852', 'gathered_video_score2_vq_reward_std': '0.54218739', 'gathered_ori_avg_reward_mean': '8.91845703', 'gathered_ori_avg_reward_std': '1.01360595'} +2026-05-30 21:16:32.194 | INFO | __main__:save_checkpoint:1917 - Checkpoint saved at step 100 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-100 +2026-05-30 21:16:32.341 | INFO | __main__:_write_timer_marker:1761 - Timer resume marker updated at step 100: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-100 +2026-05-30 21:16:34.769 | INFO | __main__:_prune_old_timer_checkpoints:1776 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-75 +2026-05-30 21:17:25.220 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 21:17:31.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:41.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_0/index_2391_seed_20199_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 21:18:56.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:56.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:13.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_0/index_2391_seed_20200_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:20:37.108 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 21:20:38.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.475 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 101 train_steps 4040 {'step': '101/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00379660', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '314.88635254', 'grpo_loop_time': '436.34625244', 'ref_model_time': '126.61508942', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88161439', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07226562', 'token_credit_video_advantage_mean': '8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.408e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00688185', 'kl_moving': '0.00688185', 'kl_stepwise': '-1.990e-08', 'gathered_avg_reward_mean': '9.07653809', 'gathered_avg_reward_std': '1.36080062', 'gathered_video_score2_mq_reward_mean': '3.19055176', 'gathered_video_score2_mq_reward_std': '0.46561784', 'gathered_video_score2_overall_reward_mean': '9.07226562', 'gathered_video_score2_overall_reward_std': '1.36193049', 'gathered_video_score2_ta_reward_mean': '2.91796875', 'gathered_video_score2_ta_reward_std': '0.56170088', 'gathered_video_score2_vq_reward_mean': '2.96801758', 'gathered_video_score2_vq_reward_std': '0.57391262', 'gathered_ori_avg_reward_mean': '9.07653809', 'gathered_ori_avg_reward_std': '1.36080062'} +2026-05-30 21:32:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 21:32:08.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:08.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:48:50.644 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-30 21:48:50.650 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-30 21:48:52.445 | INFO | __main__:print_training_configuration:963 - SYSTEM & HARDWARE CONFIGURATION +2026-05-30 21:48:52.445 | INFO | __main__:print_training_configuration:964 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.446 | INFO | __main__:print_training_configuration:965 - Number of GPUs : 64 +2026-05-30 21:48:52.446 | INFO | __main__:print_training_configuration:966 - Local rank : 0 +2026-05-30 21:48:52.447 | INFO | __main__:print_training_configuration:967 - Global rank : 0 +2026-05-30 21:48:52.447 | INFO | __main__:print_training_configuration:968 - Device : cuda:0 +2026-05-30 21:48:52.448 | INFO | __main__:print_training_configuration:969 - Master weight dtype : torch.float32 +2026-05-30 21:48:52.448 | INFO | __main__:print_training_configuration:971 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.448 | INFO | __main__:print_training_configuration:972 - MODEL PARAMETERS & ARCHITECTURE +2026-05-30 21:48:52.449 | INFO | __main__:print_training_configuration:973 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.460 | INFO | __main__:print_training_configuration:976 - Number of total : 8,326,608,160 +2026-05-30 21:48:52.461 | INFO | __main__:print_training_configuration:977 - Total trainable parameters : 8,326,608,160 +2026-05-30 21:48:52.461 | INFO | __main__:print_training_configuration:978 - DIT model root : ./ckpts +2026-05-30 21:48:52.462 | INFO | __main__:print_training_configuration:979 - Transformer version : 480p_t2v +2026-05-30 21:48:52.462 | INFO | __main__:print_training_configuration:980 - Attention mode : flash +2026-05-30 21:48:52.463 | INFO | __main__:print_training_configuration:981 - Gradient checkpointing : True +2026-05-30 21:48:52.463 | INFO | __main__:print_training_configuration:982 - FSDP enabled : True +2026-05-30 21:48:52.464 | INFO | __main__:print_training_configuration:983 - LoRA enabled : False +2026-05-30 21:48:52.464 | INFO | __main__:print_training_configuration:984 - Glyph byT5 v2 : True +2026-05-30 21:48:52.464 | INFO | __main__:print_training_configuration:985 - ByT5 max length : 256 +2026-05-30 21:48:52.465 | INFO | __main__:print_training_configuration:987 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.465 | INFO | __main__:print_training_configuration:988 - DISTRIBUTED TRAINING CONFIGURATION +2026-05-30 21:48:52.466 | INFO | __main__:print_training_configuration:989 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.466 | INFO | __main__:print_training_configuration:990 - World size : 64 +2026-05-30 21:48:52.466 | INFO | __main__:print_training_configuration:991 - DP degree : 32 +2026-05-30 21:48:52.467 | INFO | __main__:print_training_configuration:992 - DP rank : 0 +2026-05-30 21:48:52.467 | INFO | __main__:print_training_configuration:993 - SP size : 2 +2026-05-30 21:48:52.468 | INFO | __main__:print_training_configuration:994 - SP rank : 0 +2026-05-30 21:48:52.468 | INFO | __main__:print_training_configuration:996 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.468 | INFO | __main__:print_training_configuration:997 - BATCH SIZE & DATA FLOW CONFIGURATION +2026-05-30 21:48:52.469 | INFO | __main__:print_training_configuration:998 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.469 | INFO | __main__:print_training_configuration:999 - Per GPU batch size : 2 +2026-05-30 21:48:52.470 | INFO | __main__:print_training_configuration:1000 - DP degree batch size : 64 +2026-05-30 21:48:52.470 | INFO | __main__:print_training_configuration:1001 - Gradient accumulation steps : 1 +2026-05-30 21:48:52.470 | INFO | __main__:print_training_configuration:1002 - Num generations : 4 +2026-05-30 21:48:52.471 | INFO | __main__:print_training_configuration:1003 - Use same noise : False +2026-05-30 21:48:52.471 | INFO | __main__:print_training_configuration:1004 - Num workers : 4 +2026-05-30 21:48:52.471 | INFO | __main__:print_training_configuration:1005 - Prefetch factor : 2 +2026-05-30 21:48:52.472 | INFO | __main__:print_training_configuration:1007 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.472 | INFO | __main__:print_training_configuration:1008 - OPTIMIZATION CONFIGURATION +2026-05-30 21:48:52.473 | INFO | __main__:print_training_configuration:1009 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.473 | INFO | __main__:print_training_configuration:1010 - Optimizer : muon +2026-05-30 21:48:52.473 | INFO | __main__:print_training_configuration:1011 - Learning rate : 1e-05 +2026-05-30 21:48:52.474 | INFO | __main__:print_training_configuration:1012 - Weight decay : 0.01 +2026-05-30 21:48:52.474 | INFO | __main__:print_training_configuration:1013 - Max gradient norm : 1.0 +2026-05-30 21:48:52.474 | INFO | __main__:print_training_configuration:1014 - LR scheduler : constant +2026-05-30 21:48:52.475 | INFO | __main__:print_training_configuration:1015 - Warmup steps : 500 +2026-05-30 21:48:52.475 | INFO | __main__:print_training_configuration:1016 - Max training steps : 10000 +2026-05-30 21:48:52.476 | INFO | __main__:print_training_configuration:1018 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.476 | INFO | __main__:print_training_configuration:1019 - INFERENCE & VALIDATION CONFIGURATION +2026-05-30 21:48:52.476 | INFO | __main__:print_training_configuration:1020 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.477 | INFO | __main__:print_training_configuration:1021 - Validation interval : 0 +2026-05-30 21:48:52.477 | INFO | __main__:print_training_configuration:1022 - Validate at step 0 : False +2026-05-30 21:48:52.478 | INFO | __main__:print_training_configuration:1023 - Validation video length : 121 +2026-05-30 21:48:52.478 | INFO | __main__:print_training_configuration:1024 - Validation global seed : 930 +2026-05-30 21:48:52.478 | INFO | __main__:print_training_configuration:1025 - Validation inference steps : 40 +2026-05-30 21:48:52.479 | INFO | __main__:print_training_configuration:1026 - Validation aspect ratio : 16:9 +2026-05-30 21:48:52.479 | INFO | __main__:print_training_configuration:1027 - Validation fixed size : 480x864 +2026-05-30 21:48:52.480 | INFO | __main__:print_training_configuration:1028 - Validation prompt rewrite : False +2026-05-30 21:48:52.480 | INFO | __main__:print_training_configuration:1029 - Validation flow shift : 5.0 +2026-05-30 21:48:52.480 | INFO | __main__:print_training_configuration:1030 - Sampling CFG scale : 6.0 +2026-05-30 21:48:52.481 | INFO | __main__:print_training_configuration:1031 - Sampling eta : 0.5 +2026-05-30 21:48:52.481 | INFO | __main__:print_training_configuration:1032 - Sampling steps (GRPO rollout) : 20 +2026-05-30 21:48:52.481 | INFO | __main__:print_training_configuration:1033 - Token credit loss mode : video_adv +2026-05-30 21:48:52.482 | INFO | __main__:print_training_configuration:1034 - Token credit ratio mode : uniform_weighted_logprob +2026-05-30 21:48:52.482 | INFO | __main__:print_training_configuration:1035 - Token credit shuffle weights : False +2026-05-30 21:48:52.483 | INFO | __main__:print_training_configuration:1036 - Token credit source model : None +2026-05-30 21:48:52.483 | INFO | __main__:print_training_configuration:1037 - Token credit advantage source : token +2026-05-30 21:48:52.483 | INFO | __main__:print_training_configuration:1038 - Random train timestep count : False +2026-05-30 21:48:52.484 | INFO | __main__:print_training_configuration:1039 - Random train timestep range : 1-20 +2026-05-30 21:48:52.484 | INFO | __main__:print_training_configuration:1043 - SDE type : sage_grpo +2026-05-30 21:48:52.485 | INFO | __main__:print_training_configuration:1045 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.485 | INFO | __main__:print_training_configuration:1046 - GRPO & KL CONFIGURATION +2026-05-30 21:48:52.485 | INFO | __main__:print_training_configuration:1047 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.486 | INFO | __main__:print_training_configuration:1048 - Reward model : video_score2 +2026-05-30 21:48:52.486 | INFO | __main__:print_training_configuration:1049 - Reward gradient credit : True +2026-05-30 21:48:52.486 | INFO | __main__:print_training_configuration:1050 - Reward credit frames : 8 +2026-05-30 21:48:52.487 | INFO | __main__:print_training_configuration:1051 - Reward credit max questions : 1 +2026-05-30 21:48:52.487 | INFO | __main__:print_training_configuration:1052 - Reward credit metric : weighted +2026-05-30 21:48:52.488 | INFO | __main__:print_training_configuration:1063 - Reward checkpoint mode : v3 +2026-05-30 21:48:52.488 | INFO | __main__:print_training_configuration:1064 - KL weight : 1e-05 +2026-05-30 21:48:52.488 | INFO | __main__:print_training_configuration:1065 - KL coef : 1e-07 +2026-05-30 21:48:52.489 | INFO | __main__:print_training_configuration:1066 - Use moving KL : True +2026-05-30 21:48:52.489 | INFO | __main__:print_training_configuration:1067 - Update ref model step : 10 +2026-05-30 21:48:52.490 | INFO | __main__:print_training_configuration:1068 - Use dual KL : True +2026-05-30 21:48:52.490 | INFO | __main__:print_training_configuration:1069 - Dual KL moving weight : 1.0 +2026-05-30 21:48:52.490 | INFO | __main__:print_training_configuration:1070 - Dual KL step weight : 0.1 +2026-05-30 21:48:52.491 | INFO | __main__:print_training_configuration:1071 - KL compute mode : rollout_phase +2026-05-30 21:48:52.491 | INFO | __main__:print_training_configuration:1072 - Use grad balancing : True +2026-05-30 21:48:52.492 | INFO | __main__:print_training_configuration:1073 - Enable timestep permutation : True +2026-05-30 21:48:52.492 | INFO | __main__:print_training_configuration:1074 - Reference mode offload : True +2026-05-30 21:48:52.492 | INFO | __main__:print_training_configuration:1076 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.493 | INFO | __main__:print_training_configuration:1077 - LOGGING & CHECKPOINTING CONFIGURATION +2026-05-30 21:48:52.493 | INFO | __main__:print_training_configuration:1078 - -------------------------------------------------------------------------------- +2026-05-30 21:48:52.493 | INFO | __main__:print_training_configuration:1079 - Output directory : ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-30 21:48:52.494 | INFO | __main__:print_training_configuration:1080 - Checkpoints directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-30 21:48:52.494 | INFO | __main__:print_training_configuration:1081 - Checkpoint interval : 100 +2026-05-30 21:48:52.495 | INFO | __main__:print_training_configuration:1082 - Timer checkpoint interval : 25 +2026-05-30 21:48:52.495 | INFO | __main__:print_training_configuration:1083 - Auto resume from timer : True +2026-05-30 21:48:52.496 | INFO | __main__:print_training_configuration:1084 - Timer checkpoint directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer +2026-05-30 21:48:52.496 | INFO | __main__:print_training_configuration:1085 - Log interval : 1 +2026-05-30 21:48:52.497 | INFO | __main__:print_training_configuration:1086 - Resume checkpoint : None +2026-05-30 21:48:52.497 | INFO | __main__:print_training_configuration:1087 - Train CSV : ['/nfs/ywang29/Reward_finetuning/SAGE-GRPO/assets/demo_train_decomposed_binary5.csv'] +2026-05-30 21:48:52.498 | INFO | __main__:print_training_configuration:1088 - Valid CSV : assets/demo_sample.csv +2026-05-30 21:48:52.498 | INFO | __main__:print_training_configuration:1089 - Seed : 42 +2026-05-30 21:48:52.498 | INFO | __main__:print_training_configuration:1090 - WandB enabled : True +2026-05-30 21:48:52.499 | INFO | __main__:print_training_configuration:1092 - WandB project : SAGE-GRPO +2026-05-30 21:48:52.499 | INFO | __main__:print_training_configuration:1093 - WandB run name : yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-30 21:48:52.500 | INFO | __main__:__init__:951 - [Effective Config] world_size=64, rank=0, local_rank=0, sp_size=2, dp_size=32, dp_rank=0, sp_rank=0, batch_size=2, num_generations=4, grad_accum=1 +2026-05-30 21:48:52.500 | INFO | __main__:_build_video_dataset:2038 - [Effective GRPO Grouping] video_batch_size=2, dp_replicas=32, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, product=64 +2026-05-30 21:48:53.171 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:48:53.172 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:48:53.172 | INFO | __main__:_train_grpo:2121 - Starting GRPO training... +2026-05-30 21:48:53.173 | INFO | __main__:_train_grpo:2122 - Max steps: 10000 +2026-05-30 21:48:53.173 | INFO | __main__:_train_grpo:2123 - Batch size: 2 +2026-05-30 21:48:53.173 | INFO | __main__:_train_grpo:2124 - Learning rate: 1e-05 +2026-05-30 21:48:53.763 | INFO | __main__:_resolve_resume_checkpoint:1818 - Auto-resuming from timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-100 +2026-05-30 21:48:53.910 | INFO | __main__:load_checkpoint:1942 - Loading checkpoint from s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-100 +2026-05-30 21:55:01.803 | INFO | __main__:load_checkpoint:1956 - Transformer model state loaded +2026-05-30 21:55:19.234 | INFO | __main__:load_checkpoint:1971 - Optimizer state loaded +2026-05-30 21:55:19.454 | INFO | __main__:load_checkpoint:1985 - Training state loaded: global_step=100 +2026-05-30 21:55:19.457 | INFO | __main__:load_checkpoint:2001 - Checkpoint loaded successfully. Resuming from step 100 +2026-05-30 21:55:20.908 | INFO | __main__:_train_grpo:2151 - [FLAG] post_train train path -> train_one_step +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 21:55:22.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:49.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28576_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 21:57:05.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:57:05.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:18.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28577_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:58:58.040 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 21:59:08.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 0, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.486 | INFO | __main__:_log_precision_diagnostics_once:1624 - [PrecisionDiag] trainable param dtypes: ['torch.float32'] +2026-05-30 22:10:52.496 | INFO | __main__:_log_precision_diagnostics_once:1627 - [PrecisionDiag] Using composable FSDP2 mixed precision (param_dtype=bf16, reduce_dtype=float32). This path does not maintain an explicit FP32 master-weight copy for parameters. +2026-05-30 22:10:52.499 | INFO | __main__:_log_precision_diagnostics_once:1640 - [PrecisionDiag] optimizer tensor-state dtypes: ['torch.float32'] (count=2804) +2026-05-30 22:10:52.499 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 101 train_steps 40 {'step': '101/10000', 'learning_rate': '0.00001000', 'loss': '2.235e-08', 'grad_norm': '0.00303622', 'KL_weight': '0.00000396', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '353.81958008', 'grpo_loop_time': '454.02517700', 'ref_model_time': '126.75903320', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87439233', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.28955078', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.506e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.01506941', 'kl_moving': '0.01506941', 'kl_stepwise': '-2.006e-08', 'gathered_avg_reward_mean': '9.28649902', 'gathered_avg_reward_std': '0.84882671', 'gathered_video_score2_mq_reward_mean': '3.21337891', 'gathered_video_score2_mq_reward_std': '0.24321260', 'gathered_video_score2_overall_reward_mean': '9.28955078', 'gathered_video_score2_overall_reward_std': '0.84651679', 'gathered_video_score2_ta_reward_mean': '3.12475586', 'gathered_video_score2_ta_reward_std': '0.50369608', 'gathered_video_score2_vq_reward_mean': '2.94836426', 'gathered_video_score2_vq_reward_std': '0.41457951', 'gathered_ori_avg_reward_mean': '9.28649902', 'gathered_ori_avg_reward_std': '0.84882671'} +2026-05-30 22:10:52.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 22:10:52.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:12:00.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_0/index_6771_seed_33047_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 22:12:16.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:16.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:13:35.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_0/index_6771_seed_33048_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:14:02.535 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 22:14:03.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.170 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 102 train_steps 80 {'step': '102/10000', 'learning_rate': '0.00001000', 'loss': '1.001e-07', 'grad_norm': '0.00452325', 'KL_weight': '0.00000792', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '319.18994141', 'grpo_loop_time': '435.66854858', 'ref_model_time': '127.28448486', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84900832', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.68554688', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.461e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.779e-09', 'kl_loss': '0.01629138', 'kl_moving': '0.01629138', 'kl_stepwise': '-1.886e-08', 'gathered_avg_reward_mean': '8.68359375', 'gathered_avg_reward_std': '1.09136701', 'gathered_video_score2_mq_reward_mean': '3.06677246', 'gathered_video_score2_mq_reward_std': '0.33505327', 'gathered_video_score2_overall_reward_mean': '8.68554688', 'gathered_video_score2_overall_reward_std': '1.09225452', 'gathered_video_score2_ta_reward_mean': '2.92126465', 'gathered_video_score2_ta_reward_std': '0.62967354', 'gathered_video_score2_vq_reward_mean': '2.69555664', 'gathered_video_score2_vq_reward_std': '0.54566336', 'gathered_ori_avg_reward_mean': '8.68359375', 'gathered_ori_avg_reward_std': '1.09136701'} +2026-05-30 22:25:33.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 22:25:33.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:42.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_0/index_5387_seed_9522_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 22:26:55.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:55.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:14.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_0/index_5387_seed_9523_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:28:36.451 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 22:28:37.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.330 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 103 train_steps 120 {'step': '103/10000', 'learning_rate': '0.00001000', 'loss': '1.108e-07', 'grad_norm': '0.00498933', 'KL_weight': '0.00000120', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.32952881', 'grpo_loop_time': '434.91186523', 'ref_model_time': '122.15171051', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85198760', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.12841797', 'token_credit_video_advantage_mean': '9.080e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.502e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.191e-09', 'kl_loss': '0.01767710', 'kl_moving': '0.01767710', 'kl_stepwise': '-2.015e-08', 'gathered_avg_reward_mean': '9.12622070', 'gathered_avg_reward_std': '1.62042725', 'gathered_video_score2_mq_reward_mean': '3.15795898', 'gathered_video_score2_mq_reward_std': '0.50252938', 'gathered_video_score2_overall_reward_mean': '9.12841797', 'gathered_video_score2_overall_reward_std': '1.62289643', 'gathered_video_score2_ta_reward_mean': '3.15014648', 'gathered_video_score2_ta_reward_std': '0.65373909', 'gathered_video_score2_vq_reward_mean': '2.81811523', 'gathered_video_score2_vq_reward_std': '0.65374637', 'gathered_ori_avg_reward_mean': '9.12622070', 'gathered_ori_avg_reward_std': '1.62042725'} +2026-05-30 22:40:03.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 22:40:03.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_0/index_3341_seed_10098_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 22:41:29.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:29.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:42:45.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_0/index_3341_seed_10099_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:43:09.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 22:43:10.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.495 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 104 train_steps 160 {'step': '104/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.00560344', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '313.71881104', 'grpo_loop_time': '433.29003906', 'ref_model_time': '125.11378479', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.82666385', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.75488281', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.467e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.794e-09', 'kl_loss': '0.01745577', 'kl_moving': '0.01745577', 'kl_stepwise': '-1.964e-08', 'gathered_avg_reward_mean': '8.75122070', 'gathered_avg_reward_std': '1.05171764', 'gathered_video_score2_mq_reward_mean': '3.10388184', 'gathered_video_score2_mq_reward_std': '0.31107059', 'gathered_video_score2_overall_reward_mean': '8.75488281', 'gathered_video_score2_overall_reward_std': '1.05350363', 'gathered_video_score2_ta_reward_mean': '2.76489258', 'gathered_video_score2_ta_reward_std': '0.75212902', 'gathered_video_score2_vq_reward_mean': '2.88244629', 'gathered_video_score2_vq_reward_std': '0.50246477', 'gathered_ori_avg_reward_mean': '8.75122070', 'gathered_ori_avg_reward_std': '1.05171764'} +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 22:54:38.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:38.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_0/index_4150_seed_30223_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 22:56:01.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:56:01.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:25.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_0/index_4150_seed_30224_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:57:49.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 22:57:50.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:19.769 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 105 train_steps 200 {'step': '105/10000', 'learning_rate': '0.00001000', 'loss': '1.630e-08', 'grad_norm': '0.00435601', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '320.68164062', 'grpo_loop_time': '433.39529419', 'ref_model_time': '127.78190613', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.90652823', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.94384766', 'token_credit_video_advantage_mean': '7.916e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.475e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.164e-08', 'kl_loss': '0.01638576', 'kl_moving': '0.01638577', 'kl_stepwise': '-1.967e-08', 'gathered_avg_reward_mean': '8.94384766', 'gathered_avg_reward_std': '1.08777654', 'gathered_video_score2_mq_reward_mean': '3.14868164', 'gathered_video_score2_mq_reward_std': '0.36636576', 'gathered_video_score2_overall_reward_mean': '8.94384766', 'gathered_video_score2_overall_reward_std': '1.09119654', 'gathered_video_score2_ta_reward_mean': '2.94506836', 'gathered_video_score2_ta_reward_std': '0.55880535', 'gathered_video_score2_vq_reward_mean': '2.85009766', 'gathered_video_score2_vq_reward_std': '0.48713905', 'gathered_ori_avg_reward_mean': '8.94384766', 'gathered_ori_avg_reward_std': '1.08777654'} +2026-05-30 23:09:20.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 23:09:20.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_0/index_7254_seed_11671_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 23:10:44.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:44.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:11.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_0/index_7254_seed_11672_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 23:12:35.171 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:12:35.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 23:12:36.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:04.964 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 106 train_steps 240 {'step': '106/10000', 'learning_rate': '0.00001000', 'loss': '1.770e-08', 'grad_norm': '0.00522837', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '328.73492432', 'grpo_loop_time': '431.28179932', 'ref_model_time': '131.26519775', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87464523', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.39746094', 'token_credit_video_advantage_mean': '-4.075e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.446e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.066e-08', 'kl_loss': '0.01570057', 'kl_moving': '0.01570057', 'kl_stepwise': '-1.988e-08', 'gathered_avg_reward_mean': '9.39819336', 'gathered_avg_reward_std': '0.94082862', 'gathered_video_score2_mq_reward_mean': '3.24011230', 'gathered_video_score2_mq_reward_std': '0.42011997', 'gathered_video_score2_overall_reward_mean': '9.39746094', 'gathered_video_score2_overall_reward_std': '0.93714947', 'gathered_video_score2_ta_reward_mean': '3.04992676', 'gathered_video_score2_ta_reward_std': '0.55744290', 'gathered_video_score2_vq_reward_mean': '3.10815430', 'gathered_video_score2_vq_reward_std': '0.42397934', 'gathered_ori_avg_reward_mean': '9.39819336', 'gathered_ori_avg_reward_std': '0.94082862'} +2026-05-30 23:24:05.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 23:24:05.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_0/index_3113_seed_32440_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 23:25:27.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:27.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:26:47.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_0/index_3113_seed_32441_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:27:14.006 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 23:27:14.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.160 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 107 train_steps 280 {'step': '107/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00410995', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '316.17251587', 'grpo_loop_time': '435.88616943', 'ref_model_time': '124.98022461', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85740650', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.67041016', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.496e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.01625462', 'kl_moving': '0.01625462', 'kl_stepwise': '-2.015e-08', 'gathered_avg_reward_mean': '8.67114258', 'gathered_avg_reward_std': '1.39963973', 'gathered_video_score2_mq_reward_mean': '3.02661133', 'gathered_video_score2_mq_reward_std': '0.46877384', 'gathered_video_score2_overall_reward_mean': '8.67041016', 'gathered_video_score2_overall_reward_std': '1.40169537', 'gathered_video_score2_ta_reward_mean': '2.78063965', 'gathered_video_score2_ta_reward_std': '0.85949630', 'gathered_video_score2_vq_reward_mean': '2.86389160', 'gathered_video_score2_vq_reward_std': '0.45387742', 'gathered_ori_avg_reward_mean': '8.67114258', 'gathered_ori_avg_reward_std': '1.39963973'} +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 23:38:46.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_0/index_7447_seed_24478_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 23:40:10.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:40:10.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:25.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_0/index_7447_seed_24479_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:41:47.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 23:41:48.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:09.774 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 108 train_steps 320 {'step': '108/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00443858', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '306.94711304', 'grpo_loop_time': '434.38467407', 'ref_model_time': '123.38285065', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86358285', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.81494141', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.460e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.01712894', 'kl_moving': '0.01712895', 'kl_stepwise': '-1.987e-08', 'gathered_avg_reward_mean': '8.81518555', 'gathered_avg_reward_std': '1.28450692', 'gathered_video_score2_mq_reward_mean': '3.04113770', 'gathered_video_score2_mq_reward_std': '0.49790946', 'gathered_video_score2_overall_reward_mean': '8.81494141', 'gathered_video_score2_overall_reward_std': '1.28425121', 'gathered_video_score2_ta_reward_mean': '2.92211914', 'gathered_video_score2_ta_reward_std': '0.55761796', 'gathered_video_score2_vq_reward_mean': '2.85192871', 'gathered_video_score2_vq_reward_std': '0.58295047', 'gathered_ori_avg_reward_mean': '8.81518555', 'gathered_ori_avg_reward_std': '1.28450692'} +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-30 23:53:10.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_0/index_9339_seed_18934_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-30 23:54:34.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:34.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:55:52.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_0/index_9339_seed_18935_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:56:14.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-30 23:56:15.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.046 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 109 train_steps 360 {'step': '109/10000', 'learning_rate': '0.00001000', 'loss': '9.546e-09', 'grad_norm': '0.00398214', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '311.44000244', 'grpo_loop_time': '435.21487427', 'ref_model_time': '124.49021149', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84935784', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.81005859', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.389e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.630e-09', 'kl_loss': '0.01577064', 'kl_moving': '0.01577064', 'kl_stepwise': '-1.908e-08', 'gathered_avg_reward_mean': '8.80554199', 'gathered_avg_reward_std': '0.97906035', 'gathered_video_score2_mq_reward_mean': '3.11547852', 'gathered_video_score2_mq_reward_std': '0.32194725', 'gathered_video_score2_overall_reward_mean': '8.81005859', 'gathered_video_score2_overall_reward_std': '0.97784764', 'gathered_video_score2_ta_reward_mean': '2.85070801', 'gathered_video_score2_ta_reward_std': '0.57611579', 'gathered_video_score2_vq_reward_mean': '2.83935547', 'gathered_video_score2_vq_reward_std': '0.54400206', 'gathered_ori_avg_reward_mean': '8.80554199', 'gathered_ori_avg_reward_std': '0.97906041'} +2026-05-31 00:07:41.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 00:07:41.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_0/index_3943_seed_33139_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 00:09:16.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:09:16.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:24.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_0/index_3943_seed_33140_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:10:45.444 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 00:10:46.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:05.460 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 110 train_steps 400 {'step': '110/10000', 'learning_rate': '0.00001000', 'loss': '-6.054e-09', 'grad_norm': '0.00444653', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '308.51220703', 'grpo_loop_time': '432.13159180', 'ref_model_time': '122.19709015', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85648847', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.21289062', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.395e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.054e-09', 'kl_loss': '0.01582367', 'kl_moving': '0.01582367', 'kl_stepwise': '-1.949e-08', 'gathered_avg_reward_mean': '9.21105957', 'gathered_avg_reward_std': '1.21031797', 'gathered_video_score2_mq_reward_mean': '3.18969727', 'gathered_video_score2_mq_reward_std': '0.42847723', 'gathered_video_score2_overall_reward_mean': '9.21289062', 'gathered_video_score2_overall_reward_std': '1.21040654', 'gathered_video_score2_ta_reward_mean': '3.11816406', 'gathered_video_score2_ta_reward_std': '0.47229999', 'gathered_video_score2_vq_reward_mean': '2.90319824', 'gathered_video_score2_vq_reward_std': '0.53899461', 'gathered_ori_avg_reward_mean': '9.21105957', 'gathered_ori_avg_reward_std': '1.21031797'} +2026-05-31 00:22:53.979 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 00:22:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_0/index_7446_seed_6033_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 00:24:23.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:23.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:25:40.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_0/index_7446_seed_6034_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:26:04.424 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 00:26:05.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.154 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 111 train_steps 440 {'step': '111/10000', 'learning_rate': '0.00001000', 'loss': '-2.049e-08', 'grad_norm': '0.00687381', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '312.89831543', 'grpo_loop_time': '433.62799072', 'ref_model_time': '125.43041992', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84861743', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.14208984', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.480e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.049e-08', 'kl_loss': '0.00836093', 'kl_moving': '0.00836093', 'kl_stepwise': '-1.976e-08', 'gathered_avg_reward_mean': '8.13928223', 'gathered_avg_reward_std': '1.56425595', 'gathered_video_score2_mq_reward_mean': '2.87719727', 'gathered_video_score2_mq_reward_std': '0.55920029', 'gathered_video_score2_overall_reward_mean': '8.14208984', 'gathered_video_score2_overall_reward_std': '1.56731367', 'gathered_video_score2_ta_reward_mean': '2.64672852', 'gathered_video_score2_ta_reward_std': '0.69797027', 'gathered_video_score2_vq_reward_mean': '2.61535645', 'gathered_video_score2_vq_reward_std': '0.61378759', 'gathered_ori_avg_reward_mean': '8.13928223', 'gathered_ori_avg_reward_std': '1.56425595'} +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 00:37:30.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_0/index_6954_seed_18434_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 00:38:56.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:56.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:18.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_0/index_6954_seed_18435_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:40:45.457 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 00:40:46.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:06.996 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 112 train_steps 480 {'step': '112/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00514307', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '320.45404053', 'grpo_loop_time': '431.59179688', 'ref_model_time': '123.46909332', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87855786', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.76513672', 'token_credit_video_advantage_mean': '4.424e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.530e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00694690', 'kl_moving': '0.00694690', 'kl_stepwise': '-1.988e-08', 'gathered_avg_reward_mean': '8.76379395', 'gathered_avg_reward_std': '1.17718017', 'gathered_video_score2_mq_reward_mean': '3.08996582', 'gathered_video_score2_mq_reward_std': '0.40536401', 'gathered_video_score2_overall_reward_mean': '8.76513672', 'gathered_video_score2_overall_reward_std': '1.18205202', 'gathered_video_score2_ta_reward_mean': '2.70239258', 'gathered_video_score2_ta_reward_std': '0.57164490', 'gathered_video_score2_vq_reward_mean': '2.97143555', 'gathered_video_score2_vq_reward_std': '0.63124055', 'gathered_ori_avg_reward_mean': '8.76379395', 'gathered_ori_avg_reward_std': '1.17718017'} +2026-05-31 00:52:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 00:52:07.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_0/index_728_seed_22358_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 00:53:28.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:28.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:54:49.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_0/index_728_seed_22359_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:55:21.193 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 00:55:22.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.219 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 113 train_steps 520 {'step': '113/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00543170', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '319.65240479', 'grpo_loop_time': '433.63635254', 'ref_model_time': '123.23564148', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86426747', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.72900391', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.437e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.304e-08', 'kl_loss': '0.00811060', 'kl_moving': '0.00811060', 'kl_stepwise': '-2.005e-08', 'gathered_avg_reward_mean': '8.73156738', 'gathered_avg_reward_std': '1.36042428', 'gathered_video_score2_mq_reward_mean': '3.05041504', 'gathered_video_score2_mq_reward_std': '0.43003345', 'gathered_video_score2_overall_reward_mean': '8.72900391', 'gathered_video_score2_overall_reward_std': '1.35688138', 'gathered_video_score2_ta_reward_mean': '2.91845703', 'gathered_video_score2_ta_reward_std': '0.71014410', 'gathered_video_score2_vq_reward_mean': '2.76269531', 'gathered_video_score2_vq_reward_std': '0.56793219', 'gathered_ori_avg_reward_mean': '8.73156738', 'gathered_ori_avg_reward_std': '1.36042440'} +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 01:06:44.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_0/index_8365_seed_10793_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 01:08:10.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:08:10.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:26.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_0/index_8365_seed_10794_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:09:51.572 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 01:09:52.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.434 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 114 train_steps 560 {'step': '114/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00436008', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '315.80801392', 'grpo_loop_time': '433.95483398', 'ref_model_time': '126.39411926', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85786617', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.18457031', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.464e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.382e-09', 'kl_loss': '0.00672169', 'kl_moving': '0.00672169', 'kl_stepwise': '-1.968e-08', 'gathered_avg_reward_mean': '9.18518066', 'gathered_avg_reward_std': '1.15133917', 'gathered_video_score2_mq_reward_mean': '3.12109375', 'gathered_video_score2_mq_reward_std': '0.35135627', 'gathered_video_score2_overall_reward_mean': '9.18457031', 'gathered_video_score2_overall_reward_std': '1.15277028', 'gathered_video_score2_ta_reward_mean': '3.03967285', 'gathered_video_score2_ta_reward_std': '0.79635459', 'gathered_video_score2_vq_reward_mean': '3.02441406', 'gathered_video_score2_vq_reward_std': '0.44820815', 'gathered_ori_avg_reward_mean': '9.18518066', 'gathered_ori_avg_reward_std': '1.15133917'} +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 01:21:16.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_0/index_4241_seed_33057_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 01:22:41.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:41.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:23:55.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_0/index_4241_seed_33058_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:24:19.485 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 01:24:20.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.437 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 115 train_steps 600 {'step': '115/10000', 'learning_rate': '0.00001000', 'loss': '1.059e-08', 'grad_norm': '0.00387077', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '303.63253784', 'grpo_loop_time': '432.06848145', 'ref_model_time': '118.28226471', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87422901', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.82714844', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.448e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.895e-09', 'kl_loss': '0.00677431', 'kl_moving': '0.00677431', 'kl_stepwise': '-1.945e-08', 'gathered_avg_reward_mean': '8.82788086', 'gathered_avg_reward_std': '1.05979669', 'gathered_video_score2_mq_reward_mean': '3.08142090', 'gathered_video_score2_mq_reward_std': '0.32255197', 'gathered_video_score2_overall_reward_mean': '8.82714844', 'gathered_video_score2_overall_reward_std': '1.06625557', 'gathered_video_score2_ta_reward_mean': '2.88098145', 'gathered_video_score2_ta_reward_std': '0.72568613', 'gathered_video_score2_vq_reward_mean': '2.86547852', 'gathered_video_score2_vq_reward_std': '0.45891055', 'gathered_ori_avg_reward_mean': '8.82788086', 'gathered_ori_avg_reward_std': '1.05979669'} +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 01:35:36.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_0/index_6121_seed_22864_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 01:36:58.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:58.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:15.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_0/index_6121_seed_22865_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:38:37.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 01:38:38.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.244 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 116 train_steps 640 {'step': '116/10000', 'learning_rate': '0.00001000', 'loss': '-6.054e-09', 'grad_norm': '0.00475716', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '307.33465576', 'grpo_loop_time': '434.69976807', 'ref_model_time': '124.29228210', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87041497', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.82104492', 'token_credit_video_advantage_mean': '-4.424e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.383e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.054e-09', 'kl_loss': '0.00745961', 'kl_moving': '0.00745961', 'kl_stepwise': '-1.915e-08', 'gathered_avg_reward_mean': '8.82629395', 'gathered_avg_reward_std': '1.86628926', 'gathered_video_score2_mq_reward_mean': '3.04199219', 'gathered_video_score2_mq_reward_std': '0.64428335', 'gathered_video_score2_overall_reward_mean': '8.82104492', 'gathered_video_score2_overall_reward_std': '1.86338556', 'gathered_video_score2_ta_reward_mean': '2.94140625', 'gathered_video_score2_ta_reward_std': '0.72240812', 'gathered_video_score2_vq_reward_mean': '2.84289551', 'gathered_video_score2_vq_reward_std': '0.76803410', 'gathered_ori_avg_reward_mean': '8.82629395', 'gathered_ori_avg_reward_std': '1.86628926'} +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 01:50:03.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:11.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_0/index_3814_seed_28667_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 01:51:27.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:27.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:52:42.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_0/index_3814_seed_28668_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:53:06.107 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 01:53:06.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.621 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 117 train_steps 680 {'step': '117/10000', 'learning_rate': '0.00001000', 'loss': '1.001e-08', 'grad_norm': '0.00483834', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '312.77261353', 'grpo_loop_time': '431.65695190', 'ref_model_time': '127.91491699', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87527943', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.61425781', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.425e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.424e-09', 'kl_loss': '0.00734407', 'kl_moving': '0.00734407', 'kl_stepwise': '-1.865e-08', 'gathered_avg_reward_mean': '8.61596680', 'gathered_avg_reward_std': '1.03340828', 'gathered_video_score2_mq_reward_mean': '3.07055664', 'gathered_video_score2_mq_reward_std': '0.32385370', 'gathered_video_score2_overall_reward_mean': '8.61425781', 'gathered_video_score2_overall_reward_std': '1.03185976', 'gathered_video_score2_ta_reward_mean': '2.63464355', 'gathered_video_score2_ta_reward_std': '0.50370514', 'gathered_video_score2_vq_reward_mean': '2.91076660', 'gathered_video_score2_vq_reward_std': '0.49953020', 'gathered_ori_avg_reward_mean': '8.61596680', 'gathered_ori_avg_reward_std': '1.03340828'} +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 02:04:34.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_0/index_367_seed_17358_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 02:05:57.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:57.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:13.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_0/index_367_seed_17359_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 02:07:38.764 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:07:38.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 02:07:39.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 0, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:00.830 | INFO | __main__:_log_progress_step:2097 - Training Progress: update_steps 118 train_steps 720 {'step': '118/10000', 'learning_rate': '0.00001000', 'loss': '1.490e-08', 'grad_norm': '0.00463575', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '311.83816528', 'grpo_loop_time': '432.39996338', 'ref_model_time': '124.65879059', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84452856', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.63671875', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.421e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-08', 'kl_loss': '0.00757572', 'kl_moving': '0.00757572', 'kl_stepwise': '-1.904e-08', 'gathered_avg_reward_mean': '8.64111328', 'gathered_avg_reward_std': '1.23695886', 'gathered_video_score2_mq_reward_mean': '3.09472656', 'gathered_video_score2_mq_reward_std': '0.36872056', 'gathered_video_score2_overall_reward_mean': '8.63671875', 'gathered_video_score2_overall_reward_std': '1.23405969', 'gathered_video_score2_ta_reward_mean': '2.80773926', 'gathered_video_score2_ta_reward_std': '0.74380684', 'gathered_video_score2_vq_reward_mean': '2.73864746', 'gathered_video_score2_vq_reward_std': '0.55822986', 'gathered_ori_avg_reward_mean': '8.64111328', 'gathered_ori_avg_reward_std': '1.23695886'} +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-31 02:19:01.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_0/index_8800_seed_13689_dp_0_rank_0_sample_0_slot_0.mp4 +2026-05-31 02:20:25.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:25.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:21:38.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_0/index_8800_seed_13690_dp_0_rank_0_sample_1_slot_1.mp4 +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:22:01.790 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-31 02:22:02.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:15.617 | INFO | __main__:__init__:946 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-06-25 18:01:15.618 | INFO | __main__:__init__:947 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-06-25 18:01:17.686 | INFO | __main__:print_training_configuration:970 - SYSTEM & HARDWARE CONFIGURATION +2026-06-25 18:01:17.687 | INFO | __main__:print_training_configuration:971 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.687 | INFO | __main__:print_training_configuration:972 - Number of GPUs : 64 +2026-06-25 18:01:17.688 | INFO | __main__:print_training_configuration:973 - Local rank : 0 +2026-06-25 18:01:17.689 | INFO | __main__:print_training_configuration:974 - Global rank : 0 +2026-06-25 18:01:17.689 | INFO | __main__:print_training_configuration:975 - Device : cuda:0 +2026-06-25 18:01:17.690 | INFO | __main__:print_training_configuration:976 - Master weight dtype : torch.float32 +2026-06-25 18:01:17.690 | INFO | __main__:print_training_configuration:978 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.691 | INFO | __main__:print_training_configuration:979 - MODEL PARAMETERS & ARCHITECTURE +2026-06-25 18:01:17.691 | INFO | __main__:print_training_configuration:980 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.703 | INFO | __main__:print_training_configuration:983 - Number of total : 8,326,608,160 +2026-06-25 18:01:17.703 | INFO | __main__:print_training_configuration:984 - Total trainable parameters : 8,326,608,160 +2026-06-25 18:01:17.704 | INFO | __main__:print_training_configuration:985 - DIT model root : ./ckpts +2026-06-25 18:01:17.705 | INFO | __main__:print_training_configuration:986 - Transformer version : 480p_t2v +2026-06-25 18:01:17.705 | INFO | __main__:print_training_configuration:987 - Attention mode : flash +2026-06-25 18:01:17.706 | INFO | __main__:print_training_configuration:988 - Gradient checkpointing : True +2026-06-25 18:01:17.706 | INFO | __main__:print_training_configuration:989 - FSDP enabled : True +2026-06-25 18:01:17.706 | INFO | __main__:print_training_configuration:990 - LoRA enabled : False +2026-06-25 18:01:17.707 | INFO | __main__:print_training_configuration:991 - Glyph byT5 v2 : True +2026-06-25 18:01:17.707 | INFO | __main__:print_training_configuration:992 - ByT5 max length : 256 +2026-06-25 18:01:17.708 | INFO | __main__:print_training_configuration:994 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.708 | INFO | __main__:print_training_configuration:995 - DISTRIBUTED TRAINING CONFIGURATION +2026-06-25 18:01:17.709 | INFO | __main__:print_training_configuration:996 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.709 | INFO | __main__:print_training_configuration:997 - World size : 64 +2026-06-25 18:01:17.710 | INFO | __main__:print_training_configuration:998 - DP degree : 32 +2026-06-25 18:01:17.710 | INFO | __main__:print_training_configuration:999 - DP rank : 0 +2026-06-25 18:01:17.710 | INFO | __main__:print_training_configuration:1000 - SP size : 2 +2026-06-25 18:01:17.711 | INFO | __main__:print_training_configuration:1001 - SP rank : 0 +2026-06-25 18:01:17.711 | INFO | __main__:print_training_configuration:1003 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.712 | INFO | __main__:print_training_configuration:1004 - BATCH SIZE & DATA FLOW CONFIGURATION +2026-06-25 18:01:17.712 | INFO | __main__:print_training_configuration:1005 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.713 | INFO | __main__:print_training_configuration:1006 - Per GPU batch size : 2 +2026-06-25 18:01:17.713 | INFO | __main__:print_training_configuration:1007 - DP degree batch size : 64 +2026-06-25 18:01:17.713 | INFO | __main__:print_training_configuration:1008 - Gradient accumulation steps : 1 +2026-06-25 18:01:17.714 | INFO | __main__:print_training_configuration:1009 - Num generations : 4 +2026-06-25 18:01:17.714 | INFO | __main__:print_training_configuration:1010 - Use same noise : False +2026-06-25 18:01:17.715 | INFO | __main__:print_training_configuration:1011 - Num workers : 4 +2026-06-25 18:01:17.715 | INFO | __main__:print_training_configuration:1012 - Prefetch factor : 2 +2026-06-25 18:01:17.716 | INFO | __main__:print_training_configuration:1014 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.716 | INFO | __main__:print_training_configuration:1015 - OPTIMIZATION CONFIGURATION +2026-06-25 18:01:17.717 | INFO | __main__:print_training_configuration:1016 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.717 | INFO | __main__:print_training_configuration:1017 - Optimizer : muon +2026-06-25 18:01:17.717 | INFO | __main__:print_training_configuration:1018 - Learning rate : 1e-05 +2026-06-25 18:01:17.718 | INFO | __main__:print_training_configuration:1019 - Weight decay : 0.01 +2026-06-25 18:01:17.718 | INFO | __main__:print_training_configuration:1020 - Max gradient norm : 1.0 +2026-06-25 18:01:17.719 | INFO | __main__:print_training_configuration:1021 - LR scheduler : constant +2026-06-25 18:01:17.719 | INFO | __main__:print_training_configuration:1022 - Warmup steps : 500 +2026-06-25 18:01:17.720 | INFO | __main__:print_training_configuration:1023 - Max training steps : 10000 +2026-06-25 18:01:17.720 | INFO | __main__:print_training_configuration:1025 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.721 | INFO | __main__:print_training_configuration:1026 - INFERENCE & VALIDATION CONFIGURATION +2026-06-25 18:01:17.721 | INFO | __main__:print_training_configuration:1027 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.721 | INFO | __main__:print_training_configuration:1028 - Validation interval : 0 +2026-06-25 18:01:17.722 | INFO | __main__:print_training_configuration:1029 - Validate at step 0 : False +2026-06-25 18:01:17.722 | INFO | __main__:print_training_configuration:1030 - Validation video length : 121 +2026-06-25 18:01:17.723 | INFO | __main__:print_training_configuration:1031 - Validation global seed : 930 +2026-06-25 18:01:17.723 | INFO | __main__:print_training_configuration:1032 - Validation inference steps : 40 +2026-06-25 18:01:17.724 | INFO | __main__:print_training_configuration:1033 - Validation aspect ratio : 16:9 +2026-06-25 18:01:17.724 | INFO | __main__:print_training_configuration:1034 - Validation fixed size : 480x864 +2026-06-25 18:01:17.725 | INFO | __main__:print_training_configuration:1035 - Validation prompt rewrite : False +2026-06-25 18:01:17.725 | INFO | __main__:print_training_configuration:1036 - Validation flow shift : 5.0 +2026-06-25 18:01:17.725 | INFO | __main__:print_training_configuration:1037 - Sampling CFG scale : 6.0 +2026-06-25 18:01:17.726 | INFO | __main__:print_training_configuration:1038 - Sampling eta : 0.5 +2026-06-25 18:01:17.726 | INFO | __main__:print_training_configuration:1039 - Sampling steps (GRPO rollout) : 20 +2026-06-25 18:01:17.727 | INFO | __main__:print_training_configuration:1040 - Token credit loss mode : video_adv +2026-06-25 18:01:17.727 | INFO | __main__:print_training_configuration:1041 - Token credit ratio mode : uniform_weighted_logprob +2026-06-25 18:01:17.727 | INFO | __main__:print_training_configuration:1042 - Token credit shuffle weights : False +2026-06-25 18:01:17.728 | INFO | __main__:print_training_configuration:1043 - Token credit spatial granularity : frame +2026-06-25 18:01:17.728 | INFO | __main__:print_training_configuration:1044 - Token credit spatial bins : 1 +2026-06-25 18:01:17.729 | INFO | __main__:print_training_configuration:1045 - Token credit spatial window : 1 +2026-06-25 18:01:17.729 | INFO | __main__:print_training_configuration:1046 - Token credit chunk size : 1 +2026-06-25 18:01:17.730 | INFO | __main__:print_training_configuration:1047 - Token credit source model : None +2026-06-25 18:01:17.730 | INFO | __main__:print_training_configuration:1048 - Token credit advantage source : token +2026-06-25 18:01:17.731 | INFO | __main__:print_training_configuration:1049 - Token credit VLM model path : None +2026-06-25 18:01:17.731 | INFO | __main__:print_training_configuration:1050 - Token credit VLM model family : None +2026-06-25 18:01:17.732 | INFO | __main__:print_training_configuration:1051 - Random train timestep count : False +2026-06-25 18:01:17.732 | INFO | __main__:print_training_configuration:1052 - Random train timestep range : 1-20 +2026-06-25 18:01:17.732 | INFO | __main__:print_training_configuration:1056 - SDE type : sage_grpo +2026-06-25 18:01:17.733 | INFO | __main__:print_training_configuration:1058 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.733 | INFO | __main__:print_training_configuration:1059 - GRPO & KL CONFIGURATION +2026-06-25 18:01:17.734 | INFO | __main__:print_training_configuration:1060 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.734 | INFO | __main__:print_training_configuration:1061 - Reward model : video_score2 +2026-06-25 18:01:17.735 | INFO | __main__:print_training_configuration:1062 - Reward gradient credit : True +2026-06-25 18:01:17.735 | INFO | __main__:print_training_configuration:1063 - Reward credit frames : 8 +2026-06-25 18:01:17.735 | INFO | __main__:print_training_configuration:1064 - Reward credit max questions : 1 +2026-06-25 18:01:17.736 | INFO | __main__:print_training_configuration:1065 - Reward credit metric : weighted +2026-06-25 18:01:17.736 | INFO | __main__:print_training_configuration:1076 - Reward checkpoint mode : v3 +2026-06-25 18:01:17.737 | INFO | __main__:print_training_configuration:1077 - KL weight : 1e-05 +2026-06-25 18:01:17.737 | INFO | __main__:print_training_configuration:1078 - KL coef : 1e-07 +2026-06-25 18:01:17.738 | INFO | __main__:print_training_configuration:1079 - Use moving KL : True +2026-06-25 18:01:17.738 | INFO | __main__:print_training_configuration:1080 - Update ref model step : 10 +2026-06-25 18:01:17.739 | INFO | __main__:print_training_configuration:1081 - Use dual KL : True +2026-06-25 18:01:17.739 | INFO | __main__:print_training_configuration:1082 - Dual KL moving weight : 1.0 +2026-06-25 18:01:17.740 | INFO | __main__:print_training_configuration:1083 - Dual KL step weight : 0.1 +2026-06-25 18:01:17.740 | INFO | __main__:print_training_configuration:1084 - KL compute mode : rollout_phase +2026-06-25 18:01:17.740 | INFO | __main__:print_training_configuration:1085 - Use grad balancing : True +2026-06-25 18:01:17.741 | INFO | __main__:print_training_configuration:1086 - Enable timestep permutation : True +2026-06-25 18:01:17.741 | INFO | __main__:print_training_configuration:1087 - Reference mode offload : True +2026-06-25 18:01:17.742 | INFO | __main__:print_training_configuration:1089 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.742 | INFO | __main__:print_training_configuration:1090 - LOGGING & CHECKPOINTING CONFIGURATION +2026-06-25 18:01:17.743 | INFO | __main__:print_training_configuration:1091 - -------------------------------------------------------------------------------- +2026-06-25 18:01:17.743 | INFO | __main__:print_training_configuration:1092 - Output directory : ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-06-25 18:01:17.744 | INFO | __main__:print_training_configuration:1093 - Checkpoints directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-06-25 18:01:17.744 | INFO | __main__:print_training_configuration:1094 - Checkpoint interval : 100 +2026-06-25 18:01:17.744 | INFO | __main__:print_training_configuration:1095 - Timer checkpoint interval : 25 +2026-06-25 18:01:17.745 | INFO | __main__:print_training_configuration:1096 - Auto resume from timer/checkpoint : True +2026-06-25 18:01:17.746 | INFO | __main__:print_training_configuration:1097 - Timer checkpoint directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer +2026-06-25 18:01:17.746 | INFO | __main__:print_training_configuration:1098 - Log interval : 1 +2026-06-25 18:01:17.746 | INFO | __main__:print_training_configuration:1099 - Resume checkpoint : None +2026-06-25 18:01:17.747 | INFO | __main__:print_training_configuration:1100 - Train CSV : ['/nfs/ywang29/Reward_finetuning/SAGE-GRPO/assets/demo_train_decomposed_binary5.csv'] +2026-06-25 18:01:17.747 | INFO | __main__:print_training_configuration:1101 - Valid CSV : assets/demo_sample.csv +2026-06-25 18:01:17.748 | INFO | __main__:print_training_configuration:1102 - Seed : 42 +2026-06-25 18:01:17.748 | INFO | __main__:print_training_configuration:1103 - WandB enabled : True +2026-06-25 18:01:17.748 | INFO | __main__:print_training_configuration:1105 - WandB project : SAGE-GRPO +2026-06-25 18:01:17.749 | INFO | __main__:print_training_configuration:1106 - WandB run name : yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-06-25 18:01:17.749 | INFO | __main__:__init__:958 - [Effective Config] world_size=64, rank=0, local_rank=0, sp_size=2, dp_size=32, dp_rank=0, sp_rank=0, batch_size=2, num_generations=4, grad_accum=1 +2026-06-25 18:01:17.750 | INFO | __main__:_build_video_dataset:2088 - [Effective GRPO Grouping] video_batch_size=2, dp_replicas=32, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, product=64 +2026-06-25 18:01:18.425 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:18.426 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:01:18.427 | INFO | __main__:_train_grpo:2171 - Starting GRPO training... +2026-06-25 18:01:18.427 | INFO | __main__:_train_grpo:2172 - Max steps: 10000 +2026-06-25 18:01:18.428 | INFO | __main__:_train_grpo:2173 - Batch size: 2 +2026-06-25 18:01:18.428 | INFO | __main__:_train_grpo:2174 - Learning rate: 1e-05 +2026-06-25 18:01:18.964 | INFO | __main__:_resolve_resume_checkpoint:1855 - Auto-resuming from timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-100 +2026-06-25 18:01:19.103 | INFO | __main__:load_checkpoint:1992 - Loading checkpoint from s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-100 +2026-06-25 18:03:50.794 | INFO | __main__:load_checkpoint:2006 - Transformer model state loaded +2026-06-25 18:04:06.905 | INFO | __main__:load_checkpoint:2021 - Optimizer state loaded +2026-06-25 18:04:07.158 | INFO | __main__:load_checkpoint:2035 - Training state loaded: global_step=100 +2026-06-25 18:04:07.161 | INFO | __main__:load_checkpoint:2051 - Checkpoint loaded successfully. Resuming from step 100 +2026-06-25 18:04:08.841 | INFO | __main__:_train_grpo:2201 - [FLAG] post_train train path -> train_one_step +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 18:04:09.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28576_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 18:05:48.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:48.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:06:56.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28577_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:07:22.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 18:07:32.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:27.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 0, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.482 | INFO | __main__:_log_precision_diagnostics_once:1647 - [PrecisionDiag] trainable param dtypes: ['torch.float32'] +2026-06-25 18:18:41.485 | INFO | __main__:_log_precision_diagnostics_once:1650 - [PrecisionDiag] Using composable FSDP2 mixed precision (param_dtype=bf16, reduce_dtype=float32). This path does not maintain an explicit FP32 master-weight copy for parameters. +2026-06-25 18:18:41.488 | INFO | __main__:_log_precision_diagnostics_once:1663 - [PrecisionDiag] optimizer tensor-state dtypes: ['torch.float32'] (count=2804) +2026-06-25 18:18:41.489 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 101 train_steps 40 {'step': '101/10000', 'learning_rate': '0.00001000', 'loss': '2.235e-08', 'grad_norm': '0.00303995', 'KL_weight': '0.00000396', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '318.71578979', 'grpo_loop_time': '436.49023438', 'ref_model_time': '114.92919159', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87439233', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.28955078', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.506e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.01506941', 'kl_moving': '0.01506941', 'kl_stepwise': '-2.006e-08', 'gathered_avg_reward_mean': '9.28649902', 'gathered_avg_reward_std': '0.84882671', 'gathered_video_score2_mq_reward_mean': '3.21337891', 'gathered_video_score2_mq_reward_std': '0.24321260', 'gathered_video_score2_overall_reward_mean': '9.28955078', 'gathered_video_score2_overall_reward_std': '0.84651679', 'gathered_video_score2_ta_reward_mean': '3.12475586', 'gathered_video_score2_ta_reward_std': '0.50369608', 'gathered_video_score2_vq_reward_mean': '2.94836426', 'gathered_video_score2_vq_reward_std': '0.41457951', 'gathered_ori_avg_reward_mean': '9.28649902', 'gathered_ori_avg_reward_std': '0.84882671'} +2026-06-25 18:18:41.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 18:18:41.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:47.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_0/index_6771_seed_33047_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 18:19:59.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:59.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:12.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_0/index_6771_seed_33048_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 18:21:34.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:21:34.298 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 18:21:35.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.386 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 102 train_steps 80 {'step': '102/10000', 'learning_rate': '0.00001000', 'loss': '9.499e-08', 'grad_norm': '0.01126446', 'KL_weight': '0.00000792', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.13571167', 'grpo_loop_time': '419.05395508', 'ref_model_time': '113.84242249', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84664142', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.72119141', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.468e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.01622594', 'kl_moving': '0.01622594', 'kl_stepwise': '-1.948e-08', 'gathered_avg_reward_mean': '8.71679688', 'gathered_avg_reward_std': '0.95417804', 'gathered_video_score2_mq_reward_mean': '3.05004883', 'gathered_video_score2_mq_reward_std': '0.27667144', 'gathered_video_score2_overall_reward_mean': '8.72119141', 'gathered_video_score2_overall_reward_std': '0.96017987', 'gathered_video_score2_ta_reward_mean': '2.94079590', 'gathered_video_score2_ta_reward_std': '0.60106897', 'gathered_video_score2_vq_reward_mean': '2.72595215', 'gathered_video_score2_vq_reward_std': '0.53363454', 'gathered_ori_avg_reward_mean': '8.71679688', 'gathered_ori_avg_reward_std': '0.95417804'} +2026-06-25 18:32:26.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 18:32:26.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:32.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_0/index_5387_seed_9522_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 18:33:49.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:49.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:34:54.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_0/index_5387_seed_9523_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:35:14.741 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 18:35:15.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.927 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 103 train_steps 120 {'step': '103/10000', 'learning_rate': '0.00001000', 'loss': '1.099e-07', 'grad_norm': '0.00526723', 'KL_weight': '0.00000120', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.17098999', 'grpo_loop_time': '420.34082031', 'ref_model_time': '113.32475281', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89198691', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.10449219', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.491e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.024e-08', 'kl_loss': '0.01798544', 'kl_moving': '0.01798544', 'kl_stepwise': '-1.918e-08', 'gathered_avg_reward_mean': '9.10363770', 'gathered_avg_reward_std': '1.36069489', 'gathered_video_score2_mq_reward_mean': '3.12036133', 'gathered_video_score2_mq_reward_std': '0.42838874', 'gathered_video_score2_overall_reward_mean': '9.10449219', 'gathered_video_score2_overall_reward_std': '1.35709691', 'gathered_video_score2_ta_reward_mean': '3.12377930', 'gathered_video_score2_ta_reward_std': '0.58712316', 'gathered_video_score2_vq_reward_mean': '2.85949707', 'gathered_video_score2_vq_reward_std': '0.60462785', 'gathered_ori_avg_reward_mean': '9.10363770', 'gathered_ori_avg_reward_std': '1.36069489'} +2026-06-25 18:46:07.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 18:46:08.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_0/index_3341_seed_10098_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 18:47:27.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:27.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:48:38.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_0/index_3341_seed_10099_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 18:49:02.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:49:02.946 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 18:49:03.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.683 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 104 train_steps 160 {'step': '104/10000', 'learning_rate': '0.00001000', 'loss': '7.916e-09', 'grad_norm': '0.00854018', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.90246582', 'grpo_loop_time': '423.68258667', 'ref_model_time': '114.18424225', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87082499', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.80957031', 'token_credit_video_advantage_mean': '3.958e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.414e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-10', 'kl_loss': '0.01801459', 'kl_moving': '0.01801459', 'kl_stepwise': '-1.944e-08', 'gathered_avg_reward_mean': '8.80993652', 'gathered_avg_reward_std': '0.96266979', 'gathered_video_score2_mq_reward_mean': '3.14880371', 'gathered_video_score2_mq_reward_std': '0.29901773', 'gathered_video_score2_overall_reward_mean': '8.80957031', 'gathered_video_score2_overall_reward_std': '0.96568501', 'gathered_video_score2_ta_reward_mean': '2.77441406', 'gathered_video_score2_ta_reward_std': '0.69950581', 'gathered_video_score2_vq_reward_mean': '2.88671875', 'gathered_video_score2_vq_reward_std': '0.48224986', 'gathered_ori_avg_reward_mean': '8.80993652', 'gathered_ori_avg_reward_std': '0.96266979'} +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 19:00:02.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:09.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_0/index_4150_seed_30223_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 19:01:24.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:24.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:34.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_0/index_4150_seed_30224_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:02:56.817 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 19:02:57.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.137 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 105 train_steps 200 {'step': '105/10000', 'learning_rate': '0.00001000', 'loss': '-1.304e-08', 'grad_norm': '0.00467235', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.84246826', 'grpo_loop_time': '418.76782227', 'ref_model_time': '114.25781250', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88372028', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.83496094', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.456e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.397e-08', 'kl_loss': '0.01643301', 'kl_moving': '0.01643301', 'kl_stepwise': '-1.890e-08', 'gathered_avg_reward_mean': '8.83337402', 'gathered_avg_reward_std': '1.05551553', 'gathered_video_score2_mq_reward_mean': '3.05346680', 'gathered_video_score2_mq_reward_std': '0.35423037', 'gathered_video_score2_overall_reward_mean': '8.83496094', 'gathered_video_score2_overall_reward_std': '1.05208695', 'gathered_video_score2_ta_reward_mean': '2.94799805', 'gathered_video_score2_ta_reward_std': '0.60986340', 'gathered_video_score2_vq_reward_mean': '2.83190918', 'gathered_video_score2_vq_reward_std': '0.44738713', 'gathered_ori_avg_reward_mean': '8.83337402', 'gathered_ori_avg_reward_std': '1.05551553'} +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 19:13:49.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_0/index_7254_seed_11671_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 19:15:10.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:15:10.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:20.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_0/index_7254_seed_11672_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:16:41.539 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 19:16:42.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.033 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 106 train_steps 240 {'step': '106/10000', 'learning_rate': '0.00001000', 'loss': '-1.630e-09', 'grad_norm': '0.00389631', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.65322876', 'grpo_loop_time': '418.62695312', 'ref_model_time': '112.65176392', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85404539', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.32812500', 'token_credit_video_advantage_mean': '5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.426e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.355e-09', 'kl_loss': '0.01561205', 'kl_moving': '0.01561205', 'kl_stepwise': '-1.958e-08', 'gathered_avg_reward_mean': '9.32971191', 'gathered_avg_reward_std': '1.02012050', 'gathered_video_score2_mq_reward_mean': '3.20776367', 'gathered_video_score2_mq_reward_std': '0.36891505', 'gathered_video_score2_overall_reward_mean': '9.32812500', 'gathered_video_score2_overall_reward_std': '1.01514411', 'gathered_video_score2_ta_reward_mean': '3.05334473', 'gathered_video_score2_ta_reward_std': '0.56908303', 'gathered_video_score2_vq_reward_mean': '3.06860352', 'gathered_video_score2_vq_reward_std': '0.45866942', 'gathered_ori_avg_reward_mean': '9.32971191', 'gathered_ori_avg_reward_std': '1.02012050'} +2026-06-25 19:27:32.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 19:27:32.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:38.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_0/index_3113_seed_32440_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 19:28:50.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:50.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:03.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_0/index_3113_seed_32441_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:30:23.032 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 19:30:24.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.809 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 107 train_steps 280 {'step': '107/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00400413', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.63186646', 'grpo_loop_time': '421.03244019', 'ref_model_time': '112.74344635', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86347222', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.75927734', 'token_credit_video_advantage_mean': '-6.054e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.480e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.01627734', 'kl_moving': '0.01627734', 'kl_stepwise': '-2.003e-08', 'gathered_avg_reward_mean': '8.76208496', 'gathered_avg_reward_std': '1.32849050', 'gathered_video_score2_mq_reward_mean': '3.06835938', 'gathered_video_score2_mq_reward_std': '0.42625135', 'gathered_video_score2_overall_reward_mean': '8.75927734', 'gathered_video_score2_overall_reward_std': '1.32948780', 'gathered_video_score2_ta_reward_mean': '2.84204102', 'gathered_video_score2_ta_reward_std': '0.84415394', 'gathered_video_score2_vq_reward_mean': '2.85168457', 'gathered_video_score2_vq_reward_std': '0.50783223', 'gathered_ori_avg_reward_mean': '8.76208496', 'gathered_ori_avg_reward_std': '1.32849038'} +2026-06-25 19:41:16.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 19:41:16.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_0/index_7447_seed_24478_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 19:42:43.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:43.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:43:50.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_0/index_7447_seed_24479_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:44:14.823 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 19:44:15.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.597 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 108 train_steps 320 {'step': '108/10000', 'learning_rate': '0.00001000', 'loss': '2.049e-08', 'grad_norm': '0.00477628', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '293.28851318', 'grpo_loop_time': '420.50823975', 'ref_model_time': '113.77256775', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89597005', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.85644531', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.493e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.049e-08', 'kl_loss': '0.01659666', 'kl_moving': '0.01659666', 'kl_stepwise': '-2.080e-08', 'gathered_avg_reward_mean': '8.85656738', 'gathered_avg_reward_std': '1.32402813', 'gathered_video_score2_mq_reward_mean': '3.08837891', 'gathered_video_score2_mq_reward_std': '0.47271249', 'gathered_video_score2_overall_reward_mean': '8.85644531', 'gathered_video_score2_overall_reward_std': '1.32305694', 'gathered_video_score2_ta_reward_mean': '2.93078613', 'gathered_video_score2_ta_reward_std': '0.57987404', 'gathered_video_score2_vq_reward_mean': '2.83740234', 'gathered_video_score2_vq_reward_std': '0.57181865', 'gathered_ori_avg_reward_mean': '8.85656738', 'gathered_ori_avg_reward_std': '1.32402813'} +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 19:55:08.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:15.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_0/index_9339_seed_18934_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 19:56:31.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:31.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:57:41.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_0/index_9339_seed_18935_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 19:58:02.111 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:58:02.260 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 19:58:03.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.236 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 109 train_steps 360 {'step': '109/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00690088', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.66583252', 'grpo_loop_time': '423.55099487', 'ref_model_time': '114.16497040', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85326314', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.85888672', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.376e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.01610585', 'kl_moving': '0.01610586', 'kl_stepwise': '-1.880e-08', 'gathered_avg_reward_mean': '8.85778809', 'gathered_avg_reward_std': '1.10095382', 'gathered_video_score2_mq_reward_mean': '3.15429688', 'gathered_video_score2_mq_reward_std': '0.41625395', 'gathered_video_score2_overall_reward_mean': '8.85888672', 'gathered_video_score2_overall_reward_std': '1.10139620', 'gathered_video_score2_ta_reward_mean': '2.83569336', 'gathered_video_score2_ta_reward_std': '0.60457337', 'gathered_video_score2_vq_reward_mean': '2.86779785', 'gathered_video_score2_vq_reward_std': '0.53411686', 'gathered_ori_avg_reward_mean': '8.85778809', 'gathered_ori_avg_reward_std': '1.10095382'} +2026-06-25 20:09:02.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 20:09:02.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:08.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_0/index_3943_seed_33139_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 20:10:22.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:22.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:35.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_0/index_3943_seed_33140_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 20:11:59.727 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:11:59.890 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 20:12:00.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:22:54.442 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 110 train_steps 400 {'step': '110/10000', 'learning_rate': '0.00001000', 'loss': '9.034e-09', 'grad_norm': '0.00375843', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '293.66760254', 'grpo_loop_time': '420.28659058', 'ref_model_time': '114.44234467', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84704602', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.18017578', 'token_credit_video_advantage_mean': '-1.024e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.412e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.848e-09', 'kl_loss': '0.01587180', 'kl_moving': '0.01587180', 'kl_stepwise': '-1.974e-08', 'gathered_avg_reward_mean': '9.17968750', 'gathered_avg_reward_std': '1.32993114', 'gathered_video_score2_mq_reward_mean': '3.20178223', 'gathered_video_score2_mq_reward_std': '0.42955333', 'gathered_video_score2_overall_reward_mean': '9.18017578', 'gathered_video_score2_overall_reward_std': '1.33170176', 'gathered_video_score2_ta_reward_mean': '3.09960938', 'gathered_video_score2_ta_reward_std': '0.53392524', 'gathered_video_score2_vq_reward_mean': '2.87829590', 'gathered_video_score2_vq_reward_std': '0.52195197', 'gathered_ori_avg_reward_mean': '9.17968750', 'gathered_ori_avg_reward_std': '1.32993114'} +2026-06-25 20:23:30.267 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 20:23:32.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_0/index_7446_seed_6033_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 20:24:52.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:52.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:05.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_0/index_7446_seed_6034_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 20:26:25.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:26:25.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 20:26:26.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.676 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 111 train_steps 440 {'step': '111/10000', 'learning_rate': '0.00001000', 'loss': '1.653e-08', 'grad_norm': '0.00661467', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.37075806', 'grpo_loop_time': '419.47412109', 'ref_model_time': '112.66030884', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88535661', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.42773438', 'token_credit_video_advantage_mean': '-1.141e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.510e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.653e-08', 'kl_loss': '0.00829998', 'kl_moving': '0.00829998', 'kl_stepwise': '-2.070e-08', 'gathered_avg_reward_mean': '8.42480469', 'gathered_avg_reward_std': '1.34922945', 'gathered_video_score2_mq_reward_mean': '2.97326660', 'gathered_video_score2_mq_reward_std': '0.46207571', 'gathered_video_score2_overall_reward_mean': '8.42773438', 'gathered_video_score2_overall_reward_std': '1.35228348', 'gathered_video_score2_ta_reward_mean': '2.76281738', 'gathered_video_score2_ta_reward_std': '0.63897848', 'gathered_video_score2_vq_reward_mean': '2.68872070', 'gathered_video_score2_vq_reward_std': '0.58761191', 'gathered_ori_avg_reward_mean': '8.42480469', 'gathered_ori_avg_reward_std': '1.34922945'} +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 20:37:18.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_0/index_6954_seed_18434_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 20:38:41.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:41.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:39:48.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_0/index_6954_seed_18435_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:40:08.806 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 20:40:09.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.134 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 112 train_steps 480 {'step': '112/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00728999', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.53143311', 'grpo_loop_time': '421.71914673', 'ref_model_time': '112.91990662', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86967635', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.86865234', 'token_credit_video_advantage_mean': '6.985e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.524e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00698747', 'kl_moving': '0.00698748', 'kl_stepwise': '-1.962e-08', 'gathered_avg_reward_mean': '8.86975098', 'gathered_avg_reward_std': '0.99687952', 'gathered_video_score2_mq_reward_mean': '3.11425781', 'gathered_video_score2_mq_reward_std': '0.40818793', 'gathered_video_score2_overall_reward_mean': '8.86865234', 'gathered_video_score2_overall_reward_std': '0.99598163', 'gathered_video_score2_ta_reward_mean': '2.73999023', 'gathered_video_score2_ta_reward_std': '0.52300453', 'gathered_video_score2_vq_reward_mean': '3.01550293', 'gathered_video_score2_vq_reward_std': '0.56407875', 'gathered_ori_avg_reward_mean': '8.86975098', 'gathered_ori_avg_reward_std': '0.99687952'} +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 20:51:03.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:09.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_0/index_728_seed_22358_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 20:52:28.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:28.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:35.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_0/index_728_seed_22359_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:53:56.305 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 20:53:57.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.619 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 113 train_steps 520 {'step': '113/10000', 'learning_rate': '0.00001000', 'loss': '-7.916e-09', 'grad_norm': '0.00532004', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.06909180', 'grpo_loop_time': '420.78222656', 'ref_model_time': '114.40778351', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88895029', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.67236328', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.437e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.916e-09', 'kl_loss': '0.00810220', 'kl_moving': '0.00810220', 'kl_stepwise': '-2.025e-08', 'gathered_avg_reward_mean': '8.67126465', 'gathered_avg_reward_std': '1.35788572', 'gathered_video_score2_mq_reward_mean': '3.02160645', 'gathered_video_score2_mq_reward_std': '0.39799622', 'gathered_video_score2_overall_reward_mean': '8.67236328', 'gathered_video_score2_overall_reward_std': '1.35817337', 'gathered_video_score2_ta_reward_mean': '2.88427734', 'gathered_video_score2_ta_reward_std': '0.73494470', 'gathered_video_score2_vq_reward_mean': '2.76538086', 'gathered_video_score2_vq_reward_std': '0.57532871', 'gathered_ori_avg_reward_mean': '8.67126465', 'gathered_ori_avg_reward_std': '1.35788572'} +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 21:04:50.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_0/index_8365_seed_10793_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 21:06:11.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:06:11.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:23.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_0/index_8365_seed_10794_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 21:07:48.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:07:48.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 21:07:49.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.573 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 114 train_steps 560 {'step': '114/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00468944', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.71078491', 'grpo_loop_time': '420.97753906', 'ref_model_time': '113.05022430', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85683495', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.17041016', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.447e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00658391', 'kl_moving': '0.00658392', 'kl_stepwise': '-1.935e-08', 'gathered_avg_reward_mean': '9.17077637', 'gathered_avg_reward_std': '1.20031798', 'gathered_video_score2_mq_reward_mean': '3.14135742', 'gathered_video_score2_mq_reward_std': '0.39752144', 'gathered_video_score2_overall_reward_mean': '9.17041016', 'gathered_video_score2_overall_reward_std': '1.19819319', 'gathered_video_score2_ta_reward_mean': '2.98852539', 'gathered_video_score2_ta_reward_std': '0.80721259', 'gathered_video_score2_vq_reward_mean': '3.04089355', 'gathered_video_score2_vq_reward_std': '0.45753542', 'gathered_ori_avg_reward_mean': '9.17077637', 'gathered_ori_avg_reward_std': '1.20031798'} +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 21:18:43.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:49.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_0/index_4241_seed_33057_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 21:20:05.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:20:05.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:17.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_0/index_4241_seed_33058_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:21:38.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 21:21:39.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.299 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 115 train_steps 600 {'step': '115/10000', 'learning_rate': '0.00001000', 'loss': '-1.062e-08', 'grad_norm': '0.00309314', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.41482544', 'grpo_loop_time': '420.54760742', 'ref_model_time': '112.64299774', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85765535', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03417969', 'token_credit_video_advantage_mean': '-3.492e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.392e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.062e-08', 'kl_loss': '0.00656657', 'kl_moving': '0.00656657', 'kl_stepwise': '-1.886e-08', 'gathered_avg_reward_mean': '9.03771973', 'gathered_avg_reward_std': '1.05685222', 'gathered_video_score2_mq_reward_mean': '3.16748047', 'gathered_video_score2_mq_reward_std': '0.30719954', 'gathered_video_score2_overall_reward_mean': '9.03417969', 'gathered_video_score2_overall_reward_std': '1.05164087', 'gathered_video_score2_ta_reward_mean': '2.93835449', 'gathered_video_score2_ta_reward_std': '0.71481192', 'gathered_video_score2_vq_reward_mean': '2.93188477', 'gathered_video_score2_vq_reward_std': '0.48474789', 'gathered_ori_avg_reward_mean': '9.03771973', 'gathered_ori_avg_reward_std': '1.05685234'} +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 21:32:31.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:37.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_0/index_6121_seed_22864_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 21:33:51.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:51.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:01.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_0/index_6121_seed_22865_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 21:35:20.628 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:35:20.758 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 21:35:21.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.307 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 116 train_steps 640 {'step': '116/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00465423', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.63827515', 'grpo_loop_time': '420.86529541', 'ref_model_time': '113.70160675', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87309980', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07031250', 'token_credit_video_advantage_mean': '-6.985e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.405e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00740845', 'kl_moving': '0.00740845', 'kl_stepwise': '-1.972e-08', 'gathered_avg_reward_mean': '9.07043457', 'gathered_avg_reward_std': '1.67905688', 'gathered_video_score2_mq_reward_mean': '3.16455078', 'gathered_video_score2_mq_reward_std': '0.58220798', 'gathered_video_score2_overall_reward_mean': '9.07031250', 'gathered_video_score2_overall_reward_std': '1.68154871', 'gathered_video_score2_ta_reward_mean': '3.04211426', 'gathered_video_score2_ta_reward_std': '0.63050950', 'gathered_video_score2_vq_reward_mean': '2.86376953', 'gathered_video_score2_vq_reward_std': '0.76153702', 'gathered_ori_avg_reward_mean': '9.07043457', 'gathered_ori_avg_reward_std': '1.67905688'} +2026-06-25 21:46:16.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 21:46:16.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:22.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_0/index_3814_seed_28667_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 21:47:36.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:36.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:48:47.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_0/index_3814_seed_28668_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:49:09.971 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 21:49:10.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.551 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 117 train_steps 680 {'step': '117/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00450088', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.74844360', 'grpo_loop_time': '423.05545044', 'ref_model_time': '112.68354034', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86590356', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.56835938', 'token_credit_video_advantage_mean': '-4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.478e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.00714378', 'kl_moving': '0.00714378', 'kl_stepwise': '-1.832e-08', 'gathered_avg_reward_mean': '8.56738281', 'gathered_avg_reward_std': '0.95522642', 'gathered_video_score2_mq_reward_mean': '3.00805664', 'gathered_video_score2_mq_reward_std': '0.27781090', 'gathered_video_score2_overall_reward_mean': '8.56835938', 'gathered_video_score2_overall_reward_std': '0.95547348', 'gathered_video_score2_ta_reward_mean': '2.61645508', 'gathered_video_score2_ta_reward_std': '0.48965377', 'gathered_video_score2_vq_reward_mean': '2.94287109', 'gathered_video_score2_vq_reward_std': '0.52673554', 'gathered_ori_avg_reward_mean': '8.56738281', 'gathered_ori_avg_reward_std': '0.95522642'} +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 22:00:05.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_0/index_367_seed_17358_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 22:01:24.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:24.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:36.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_0/index_367_seed_17359_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:02:57.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 22:02:58.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.471 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 118 train_steps 720 {'step': '118/10000', 'learning_rate': '0.00001000', 'loss': '8.382e-09', 'grad_norm': '0.00382227', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.28485107', 'grpo_loop_time': '418.84527588', 'ref_model_time': '114.48852539', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87455487', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.56738281', 'token_credit_video_advantage_mean': '-2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.386e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00724955', 'kl_moving': '0.00724956', 'kl_stepwise': '-1.918e-08', 'gathered_avg_reward_mean': '8.56689453', 'gathered_avg_reward_std': '1.35992551', 'gathered_video_score2_mq_reward_mean': '3.06909180', 'gathered_video_score2_mq_reward_std': '0.45604992', 'gathered_video_score2_overall_reward_mean': '8.56738281', 'gathered_video_score2_overall_reward_std': '1.36145246', 'gathered_video_score2_ta_reward_mean': '2.77026367', 'gathered_video_score2_ta_reward_std': '0.74951118', 'gathered_video_score2_vq_reward_mean': '2.72753906', 'gathered_video_score2_vq_reward_std': '0.55346632', 'gathered_ori_avg_reward_mean': '8.56689453', 'gathered_ori_avg_reward_std': '1.35992551'} +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 22:13:50.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_0/index_8800_seed_13689_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 22:15:13.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:15:13.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:22.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_0/index_8800_seed_13690_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:16:44.839 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 22:16:45.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.308 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 119 train_steps 760 {'step': '119/10000', 'learning_rate': '0.00001000', 'loss': '-3.492e-10', 'grad_norm': '0.00447779', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.53280640', 'grpo_loop_time': '420.80776978', 'ref_model_time': '112.62687683', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88071722', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.05224609', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.428e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.048e-09', 'kl_loss': '0.00733185', 'kl_moving': '0.00733185', 'kl_stepwise': '-2.026e-08', 'gathered_avg_reward_mean': '9.04797363', 'gathered_avg_reward_std': '1.19490302', 'gathered_video_score2_mq_reward_mean': '3.17626953', 'gathered_video_score2_mq_reward_std': '0.39911160', 'gathered_video_score2_overall_reward_mean': '9.05224609', 'gathered_video_score2_overall_reward_std': '1.19773614', 'gathered_video_score2_ta_reward_mean': '2.96496582', 'gathered_video_score2_ta_reward_std': '0.62299025', 'gathered_video_score2_vq_reward_mean': '2.90673828', 'gathered_video_score2_vq_reward_std': '0.54235166', 'gathered_ori_avg_reward_mean': '9.04797363', 'gathered_ori_avg_reward_std': '1.19490302'} +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 22:27:39.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:45.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_0/index_657_seed_28953_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 22:29:02.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:29:02.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:11.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_0/index_657_seed_28954_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:30:31.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 22:30:32.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:22.659 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 120 train_steps 800 {'step': '120/10000', 'learning_rate': '0.00001000', 'loss': '1.676e-08', 'grad_norm': '0.00428806', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.38751221', 'grpo_loop_time': '418.48681641', 'ref_model_time': '112.91280365', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86116183', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.84033203', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.474e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.676e-08', 'kl_loss': '0.00794348', 'kl_moving': '0.00794348', 'kl_stepwise': '-1.945e-08', 'gathered_avg_reward_mean': '8.83801270', 'gathered_avg_reward_std': '0.98020822', 'gathered_video_score2_mq_reward_mean': '3.10058594', 'gathered_video_score2_mq_reward_std': '0.36525917', 'gathered_video_score2_overall_reward_mean': '8.84033203', 'gathered_video_score2_overall_reward_std': '0.98130280', 'gathered_video_score2_ta_reward_mean': '2.87756348', 'gathered_video_score2_ta_reward_std': '0.49648780', 'gathered_video_score2_vq_reward_mean': '2.85986328', 'gathered_video_score2_vq_reward_std': '0.52627367', 'gathered_ori_avg_reward_mean': '8.83801270', 'gathered_ori_avg_reward_std': '0.98020828'} +2026-06-25 22:41:57.946 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 22:41:59.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:41:59.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:07.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_0/index_5415_seed_11619_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 22:43:24.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:24.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:31.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_0/index_5415_seed_11620_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:44:51.463 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 22:44:52.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.063 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.544 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 121 train_steps 840 {'step': '121/10000', 'learning_rate': '0.00001000', 'loss': '-1.490e-08', 'grad_norm': '0.00352763', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.53082275', 'grpo_loop_time': '420.84179688', 'ref_model_time': '112.30456543', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83263940', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.72363281', 'token_credit_video_advantage_mean': '-9.080e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.416e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.490e-08', 'kl_loss': '0.00679415', 'kl_moving': '0.00679415', 'kl_stepwise': '-1.928e-08', 'gathered_avg_reward_mean': '8.72790527', 'gathered_avg_reward_std': '1.21284068', 'gathered_video_score2_mq_reward_mean': '3.03112793', 'gathered_video_score2_mq_reward_std': '0.37621996', 'gathered_video_score2_overall_reward_mean': '8.72363281', 'gathered_video_score2_overall_reward_std': '1.21088982', 'gathered_video_score2_ta_reward_mean': '2.94519043', 'gathered_video_score2_ta_reward_std': '0.66892534', 'gathered_video_score2_vq_reward_mean': '2.75158691', 'gathered_video_score2_vq_reward_std': '0.52845871', 'gathered_ori_avg_reward_mean': '8.72790527', 'gathered_ori_avg_reward_std': '1.21284080'} +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 22:55:45.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:51.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_0/index_6892_seed_11576_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 22:57:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:57:07.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:16.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_0/index_6892_seed_11577_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:58:38.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 22:58:39.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:29.999 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 122 train_steps 880 {'step': '122/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00666605', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.39306641', 'grpo_loop_time': '418.72781372', 'ref_model_time': '112.86054993', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84352958', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.71240234', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.418e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00698660', 'kl_moving': '0.00698660', 'kl_stepwise': '-1.917e-08', 'gathered_avg_reward_mean': '8.70812988', 'gathered_avg_reward_std': '1.12676728', 'gathered_video_score2_mq_reward_mean': '3.00744629', 'gathered_video_score2_mq_reward_std': '0.44357750', 'gathered_video_score2_overall_reward_mean': '8.71240234', 'gathered_video_score2_overall_reward_std': '1.12748182', 'gathered_video_score2_ta_reward_mean': '2.86120605', 'gathered_video_score2_ta_reward_std': '0.63004369', 'gathered_video_score2_vq_reward_mean': '2.83947754', 'gathered_video_score2_vq_reward_std': '0.56179076', 'gathered_ori_avg_reward_mean': '8.70812988', 'gathered_ori_avg_reward_std': '1.12676728'} +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 23:09:30.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_0/index_4192_seed_17251_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 23:10:49.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:49.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:01.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_0/index_4192_seed_17252_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 23:12:21.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:12:21.704 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 23:12:22.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.880 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 123 train_steps 920 {'step': '123/10000', 'learning_rate': '0.00001000', 'loss': '1.024e-09', 'grad_norm': '0.00384963', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.59136963', 'grpo_loop_time': '419.08319092', 'ref_model_time': '112.95431519', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.82726693', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.51318359', 'token_credit_video_advantage_mean': '-6.054e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.391e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.024e-09', 'kl_loss': '0.00681987', 'kl_moving': '0.00681988', 'kl_stepwise': '-1.970e-08', 'gathered_avg_reward_mean': '9.51293945', 'gathered_avg_reward_std': '1.21288598', 'gathered_video_score2_mq_reward_mean': '3.31140137', 'gathered_video_score2_mq_reward_std': '0.46887586', 'gathered_video_score2_overall_reward_mean': '9.51318359', 'gathered_video_score2_overall_reward_std': '1.21416914', 'gathered_video_score2_ta_reward_mean': '3.14038086', 'gathered_video_score2_ta_reward_std': '0.63434327', 'gathered_video_score2_vq_reward_mean': '3.06115723', 'gathered_video_score2_vq_reward_std': '0.43536192', 'gathered_ori_avg_reward_mean': '9.51293945', 'gathered_ori_avg_reward_std': '1.21288610'} +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 23:23:15.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_0/index_7474_seed_23209_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 23:24:38.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:38.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:25:47.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_0/index_7474_seed_23210_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:26:10.267 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 23:26:11.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.022 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 124 train_steps 960 {'step': '124/10000', 'learning_rate': '0.00001000', 'loss': '4.191e-09', 'grad_norm': '0.00577639', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.83566284', 'grpo_loop_time': '421.45709229', 'ref_model_time': '113.95275879', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89162189', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20654297', 'token_credit_video_advantage_mean': '2.328e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.384e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00730777', 'kl_moving': '0.00730777', 'kl_stepwise': '-1.918e-08', 'gathered_avg_reward_mean': '9.20556641', 'gathered_avg_reward_std': '1.02760351', 'gathered_video_score2_mq_reward_mean': '3.23022461', 'gathered_video_score2_mq_reward_std': '0.35135892', 'gathered_video_score2_overall_reward_mean': '9.20654297', 'gathered_video_score2_overall_reward_std': '1.03359425', 'gathered_video_score2_ta_reward_mean': '2.99377441', 'gathered_video_score2_ta_reward_std': '0.59185839', 'gathered_video_score2_vq_reward_mean': '2.98156738', 'gathered_video_score2_vq_reward_std': '0.55401719', 'gathered_ori_avg_reward_mean': '9.20556641', 'gathered_ori_avg_reward_std': '1.02760351'} +2026-06-25 23:37:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 23:37:07.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:13.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_0/index_9364_seed_12316_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 23:38:29.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:29.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:39:38.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_0/index_9364_seed_12317_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:40:01.116 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 23:40:02.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:50:54.175 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 125 train_steps 1000 {'step': '125/10000', 'learning_rate': '0.00001000', 'loss': '-1.211e-08', 'grad_norm': '0.00437919', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.80786133', 'grpo_loop_time': '419.40768433', 'ref_model_time': '113.95996094', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84609616', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.80664062', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.390e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.211e-08', 'kl_loss': '0.00709591', 'kl_moving': '0.00709591', 'kl_stepwise': '-1.832e-08', 'gathered_avg_reward_mean': '8.80712891', 'gathered_avg_reward_std': '0.89354545', 'gathered_video_score2_mq_reward_mean': '3.08740234', 'gathered_video_score2_mq_reward_std': '0.26901728', 'gathered_video_score2_overall_reward_mean': '8.80664062', 'gathered_video_score2_overall_reward_std': '0.89545828', 'gathered_video_score2_ta_reward_mean': '2.98681641', 'gathered_video_score2_ta_reward_std': '0.45915446', 'gathered_video_score2_vq_reward_mean': '2.73291016', 'gathered_video_score2_vq_reward_std': '0.49232092', 'gathered_ori_avg_reward_mean': '8.80712891', 'gathered_ori_avg_reward_std': '0.89354551'} +2026-06-25 23:51:51.668 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 125 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-125 +2026-06-25 23:51:51.805 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 125: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-125 +2026-06-25 23:51:51.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-25 23:51:52.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:58.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_0/index_1319_seed_27710_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-25 23:53:12.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:53:12.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:23.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_0/index_1319_seed_27711_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:54:43.096 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-25 23:54:43.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.319 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 126 train_steps 1040 {'step': '126/10000', 'learning_rate': '0.00001000', 'loss': '5.355e-09', 'grad_norm': '0.00562827', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.44406128', 'grpo_loop_time': '419.54040527', 'ref_model_time': '112.76136780', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89802229', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.84326172', 'token_credit_video_advantage_mean': '-1.164e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.476e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.424e-09', 'kl_loss': '0.00710776', 'kl_moving': '0.00710776', 'kl_stepwise': '-2.038e-08', 'gathered_avg_reward_mean': '8.84655762', 'gathered_avg_reward_std': '1.34814894', 'gathered_video_score2_mq_reward_mean': '3.10681152', 'gathered_video_score2_mq_reward_std': '0.45593035', 'gathered_video_score2_overall_reward_mean': '8.84326172', 'gathered_video_score2_overall_reward_std': '1.34580934', 'gathered_video_score2_ta_reward_mean': '2.81115723', 'gathered_video_score2_ta_reward_std': '0.59987903', 'gathered_video_score2_vq_reward_mean': '2.92858887', 'gathered_video_score2_vq_reward_std': '0.55561495', 'gathered_ori_avg_reward_mean': '8.84655762', 'gathered_ori_avg_reward_std': '1.34814894'} +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 00:05:36.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:42.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_0/index_3260_seed_679_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 00:06:57.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:57.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:07.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_0/index_3260_seed_680_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:08:27.439 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 00:08:28.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.012 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 127 train_steps 1080 {'step': '127/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00422704', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.76876831', 'grpo_loop_time': '419.64453125', 'ref_model_time': '113.17579651', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87407720', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.76904297', 'token_credit_video_advantage_mean': '-1.118e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.420e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00704024', 'kl_moving': '0.00704024', 'kl_stepwise': '-1.925e-08', 'gathered_avg_reward_mean': '8.76831055', 'gathered_avg_reward_std': '1.06984651', 'gathered_video_score2_mq_reward_mean': '3.07873535', 'gathered_video_score2_mq_reward_std': '0.30406809', 'gathered_video_score2_overall_reward_mean': '8.76904297', 'gathered_video_score2_overall_reward_std': '1.07185984', 'gathered_video_score2_ta_reward_mean': '2.81420898', 'gathered_video_score2_ta_reward_std': '0.66994673', 'gathered_video_score2_vq_reward_mean': '2.87536621', 'gathered_video_score2_vq_reward_std': '0.54049790', 'gathered_ori_avg_reward_mean': '8.76831055', 'gathered_ori_avg_reward_std': '1.06984663'} +2026-06-26 00:19:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 00:19:21.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_0/index_2884_seed_11115_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 00:20:42.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:42.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:21:50.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_0/index_2884_seed_11116_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:22:10.531 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 00:22:11.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.360 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 128 train_steps 1120 {'step': '128/10000', 'learning_rate': '0.00001000', 'loss': '1.676e-09', 'grad_norm': '0.00704696', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.57858276', 'grpo_loop_time': '421.51068115', 'ref_model_time': '113.52445984', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87368751', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01660156', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.414e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.774e-09', 'kl_loss': '0.00850071', 'kl_moving': '0.00850071', 'kl_stepwise': '-1.910e-08', 'gathered_avg_reward_mean': '9.01831055', 'gathered_avg_reward_std': '1.37120020', 'gathered_video_score2_mq_reward_mean': '3.16967773', 'gathered_video_score2_mq_reward_std': '0.43287057', 'gathered_video_score2_overall_reward_mean': '9.01660156', 'gathered_video_score2_overall_reward_std': '1.37134373', 'gathered_video_score2_ta_reward_mean': '3.09008789', 'gathered_video_score2_ta_reward_std': '0.52492023', 'gathered_video_score2_vq_reward_mean': '2.75854492', 'gathered_video_score2_vq_reward_std': '0.64518601', 'gathered_ori_avg_reward_mean': '9.01831055', 'gathered_ori_avg_reward_std': '1.37120032'} +2026-06-26 00:33:07.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 00:33:07.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:13.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_0/index_7718_seed_23896_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 00:34:26.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:26.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:35:37.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_0/index_7718_seed_23897_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 00:36:02.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:36:03.136 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 00:36:04.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:02.940 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 129 train_steps 1160 {'step': '129/10000', 'learning_rate': '0.00001000', 'loss': '-5.821e-09', 'grad_norm': '0.00660073', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.80303955', 'grpo_loop_time': '423.04638672', 'ref_model_time': '114.21837616', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88491195', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.08691406', 'token_credit_video_advantage_mean': '3.492e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.821e-09', 'kl_loss': '0.00733304', 'kl_moving': '0.00733304', 'kl_stepwise': '-1.909e-08', 'gathered_avg_reward_mean': '9.08532715', 'gathered_avg_reward_std': '1.16342187', 'gathered_video_score2_mq_reward_mean': '3.19311523', 'gathered_video_score2_mq_reward_std': '0.39320794', 'gathered_video_score2_overall_reward_mean': '9.08691406', 'gathered_video_score2_overall_reward_std': '1.16284895', 'gathered_video_score2_ta_reward_mean': '2.99877930', 'gathered_video_score2_ta_reward_std': '0.51812363', 'gathered_video_score2_vq_reward_mean': '2.89343262', 'gathered_video_score2_vq_reward_std': '0.57307059', 'gathered_ori_avg_reward_mean': '9.08532715', 'gathered_ori_avg_reward_std': '1.16342187'} +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 00:47:03.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:09.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_0/index_6763_seed_10134_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 00:48:23.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:23.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:37.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_0/index_6763_seed_10135_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:50:00.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 00:50:00.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:00:53.157 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 130 train_steps 1200 {'step': '130/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00882594', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.38720703', 'grpo_loop_time': '419.92459106', 'ref_model_time': '113.59909821', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88300258', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.40966797', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.417e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00745736', 'kl_moving': '0.00745736', 'kl_stepwise': '-1.877e-08', 'gathered_avg_reward_mean': '9.40942383', 'gathered_avg_reward_std': '1.39331055', 'gathered_video_score2_mq_reward_mean': '3.29675293', 'gathered_video_score2_mq_reward_std': '0.53857136', 'gathered_video_score2_overall_reward_mean': '9.40966797', 'gathered_video_score2_overall_reward_std': '1.39383757', 'gathered_video_score2_ta_reward_mean': '2.94799805', 'gathered_video_score2_ta_reward_std': '0.67011291', 'gathered_video_score2_vq_reward_mean': '3.16467285', 'gathered_video_score2_vq_reward_std': '0.49775633', 'gathered_ori_avg_reward_mean': '9.40942383', 'gathered_ori_avg_reward_std': '1.39331055'} +2026-06-26 01:01:28.731 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 01:01:30.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:38.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_0/index_8075_seed_1736_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 01:02:52.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:52.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:03.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_0/index_8075_seed_1737_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:04:24.921 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 01:04:25.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.752 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 131 train_steps 1240 {'step': '131/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-10', 'grad_norm': '0.00611770', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.34295654', 'grpo_loop_time': '418.74902344', 'ref_model_time': '112.47180939', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87165946', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.88281250', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.434e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00681395', 'kl_moving': '0.00681395', 'kl_stepwise': '-2.009e-08', 'gathered_avg_reward_mean': '8.87756348', 'gathered_avg_reward_std': '1.07237422', 'gathered_video_score2_mq_reward_mean': '3.15100098', 'gathered_video_score2_mq_reward_std': '0.36755627', 'gathered_video_score2_overall_reward_mean': '8.88281250', 'gathered_video_score2_overall_reward_std': '1.07348692', 'gathered_video_score2_ta_reward_mean': '2.90087891', 'gathered_video_score2_ta_reward_std': '0.55438459', 'gathered_video_score2_vq_reward_mean': '2.82568359', 'gathered_video_score2_vq_reward_std': '0.55713534', 'gathered_ori_avg_reward_mean': '8.87756348', 'gathered_ori_avg_reward_std': '1.07237422'} +2026-06-26 01:15:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 01:15:17.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:17.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_0/index_9874_seed_23792_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 01:16:41.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:41.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:17:50.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_0/index_9874_seed_23793_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:18:11.186 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 01:18:11.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.407 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 132 train_steps 1280 {'step': '132/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00388266', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.00390625', 'grpo_loop_time': '421.18627930', 'ref_model_time': '114.00210571', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86353749', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.61669922', 'token_credit_video_advantage_mean': '-4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.397e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00655967', 'kl_moving': '0.00655967', 'kl_stepwise': '-1.910e-08', 'gathered_avg_reward_mean': '9.61389160', 'gathered_avg_reward_std': '1.19922984', 'gathered_video_score2_mq_reward_mean': '3.25854492', 'gathered_video_score2_mq_reward_std': '0.44824904', 'gathered_video_score2_overall_reward_mean': '9.61669922', 'gathered_video_score2_overall_reward_std': '1.19541776', 'gathered_video_score2_ta_reward_mean': '3.28125000', 'gathered_video_score2_ta_reward_std': '0.56498337', 'gathered_video_score2_vq_reward_mean': '3.07409668', 'gathered_video_score2_vq_reward_std': '0.48452067', 'gathered_ori_avg_reward_mean': '9.61389160', 'gathered_ori_avg_reward_std': '1.19922972'} +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 01:29:08.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:15.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_0/index_5918_seed_27276_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 01:30:31.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:31.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:31:40.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_0/index_5918_seed_27277_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:32:02.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 01:32:03.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.063 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 133 train_steps 1320 {'step': '133/10000', 'learning_rate': '0.00001000', 'loss': '7.683e-09', 'grad_norm': '0.00626005', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.60369873', 'grpo_loop_time': '420.34246826', 'ref_model_time': '114.16038513', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85672283', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.55322266', 'token_credit_video_advantage_mean': '-1.397e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.479e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.821e-09', 'kl_loss': '0.00689281', 'kl_moving': '0.00689282', 'kl_stepwise': '-1.940e-08', 'gathered_avg_reward_mean': '9.55285645', 'gathered_avg_reward_std': '1.17002594', 'gathered_video_score2_mq_reward_mean': '3.29125977', 'gathered_video_score2_mq_reward_std': '0.40809909', 'gathered_video_score2_overall_reward_mean': '9.55322266', 'gathered_video_score2_overall_reward_std': '1.16986394', 'gathered_video_score2_ta_reward_mean': '3.24328613', 'gathered_video_score2_ta_reward_std': '0.56260657', 'gathered_video_score2_vq_reward_mean': '3.01831055', 'gathered_video_score2_vq_reward_std': '0.51693690', 'gathered_ori_avg_reward_mean': '9.55285645', 'gathered_ori_avg_reward_std': '1.17002606'} +2026-06-26 01:42:58.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 01:42:58.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:04.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_0/index_113_seed_5468_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 01:44:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:18.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:29.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_0/index_113_seed_5469_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:45:49.783 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 01:45:50.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.012 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 134 train_steps 1360 {'step': '134/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-10', 'grad_norm': '0.00786667', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.70431519', 'grpo_loop_time': '420.26214600', 'ref_model_time': '112.46578217', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86480880', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.94775391', 'token_credit_video_advantage_mean': '5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.434e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-10', 'kl_loss': '0.00732388', 'kl_moving': '0.00732388', 'kl_stepwise': '-1.971e-08', 'gathered_avg_reward_mean': '8.94580078', 'gathered_avg_reward_std': '1.03654909', 'gathered_video_score2_mq_reward_mean': '3.16845703', 'gathered_video_score2_mq_reward_std': '0.39881521', 'gathered_video_score2_overall_reward_mean': '8.94775391', 'gathered_video_score2_overall_reward_std': '1.03963590', 'gathered_video_score2_ta_reward_mean': '2.78540039', 'gathered_video_score2_ta_reward_std': '0.51291752', 'gathered_video_score2_vq_reward_mean': '2.99194336', 'gathered_video_score2_vq_reward_std': '0.51727951', 'gathered_ori_avg_reward_mean': '8.94580078', 'gathered_ori_avg_reward_std': '1.03654909'} +2026-06-26 01:56:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 01:56:44.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:50.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_0/index_776_seed_23861_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 01:58:06.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:58:06.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:17.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_0/index_776_seed_23862_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 01:59:39.165 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:59:39.330 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 01:59:40.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.827 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 135 train_steps 1400 {'step': '135/10000', 'learning_rate': '0.00001000', 'loss': '1.583e-08', 'grad_norm': '0.00846814', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.81262207', 'grpo_loop_time': '419.22338867', 'ref_model_time': '112.97960663', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85730535', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.15576172', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.439e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.583e-08', 'kl_loss': '0.00758890', 'kl_moving': '0.00758890', 'kl_stepwise': '-1.919e-08', 'gathered_avg_reward_mean': '9.15881348', 'gathered_avg_reward_std': '1.43422306', 'gathered_video_score2_mq_reward_mean': '3.17626953', 'gathered_video_score2_mq_reward_std': '0.41278425', 'gathered_video_score2_overall_reward_mean': '9.15576172', 'gathered_video_score2_overall_reward_std': '1.43380654', 'gathered_video_score2_ta_reward_mean': '2.99816895', 'gathered_video_score2_ta_reward_std': '0.77255630', 'gathered_video_score2_vq_reward_mean': '2.98437500', 'gathered_video_score2_vq_reward_std': '0.57957155', 'gathered_ori_avg_reward_mean': '9.15881348', 'gathered_ori_avg_reward_std': '1.43422306'} +2026-06-26 02:10:30.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 02:10:30.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:37.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_0/index_5179_seed_11323_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 02:11:52.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:52.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:03.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_0/index_5179_seed_11324_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:13:23.911 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 02:13:24.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.831 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 136 train_steps 1440 {'step': '136/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-10', 'grad_norm': '0.00424229', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.47656250', 'grpo_loop_time': '419.70288086', 'ref_model_time': '112.62017822', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86555243', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.77685547', 'token_credit_video_advantage_mean': '3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-10', 'kl_loss': '0.00726837', 'kl_moving': '0.00726837', 'kl_stepwise': '-2.024e-08', 'gathered_avg_reward_mean': '8.77478027', 'gathered_avg_reward_std': '1.32323003', 'gathered_video_score2_mq_reward_mean': '3.04650879', 'gathered_video_score2_mq_reward_std': '0.37457868', 'gathered_video_score2_overall_reward_mean': '8.77685547', 'gathered_video_score2_overall_reward_std': '1.32324314', 'gathered_video_score2_ta_reward_mean': '2.92419434', 'gathered_video_score2_ta_reward_std': '0.59377038', 'gathered_video_score2_vq_reward_mean': '2.80407715', 'gathered_video_score2_vq_reward_std': '0.62716043', 'gathered_ori_avg_reward_mean': '8.77478027', 'gathered_ori_avg_reward_std': '1.32323003'} +2026-06-26 02:24:15.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 02:24:15.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:15.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_0/index_5724_seed_12245_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 02:25:40.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:40.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:26:46.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_0/index_5724_seed_12246_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:27:11.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 02:27:11.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.927 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 137 train_steps 1480 {'step': '137/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00527417', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.06149292', 'grpo_loop_time': '420.93945312', 'ref_model_time': '114.28649902', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88816309', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.21533203', 'token_credit_video_advantage_mean': '6.054e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.367e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00745101', 'kl_moving': '0.00745102', 'kl_stepwise': '-1.891e-08', 'gathered_avg_reward_mean': '9.21972656', 'gathered_avg_reward_std': '1.20528901', 'gathered_video_score2_mq_reward_mean': '3.19165039', 'gathered_video_score2_mq_reward_std': '0.39522913', 'gathered_video_score2_overall_reward_mean': '9.21533203', 'gathered_video_score2_overall_reward_std': '1.19837379', 'gathered_video_score2_ta_reward_mean': '3.20825195', 'gathered_video_score2_ta_reward_std': '0.61886829', 'gathered_video_score2_vq_reward_mean': '2.81982422', 'gathered_video_score2_vq_reward_std': '0.52768701', 'gathered_ori_avg_reward_mean': '9.21972656', 'gathered_ori_avg_reward_std': '1.20528901'} +2026-06-26 02:38:07.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 02:38:08.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_0/index_6980_seed_10946_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 02:39:29.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:29.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:37.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_0/index_6980_seed_10947_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 02:40:59.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:40:59.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 02:41:00.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.719 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 138 train_steps 1520 {'step': '138/10000', 'learning_rate': '0.00001000', 'loss': '1.397e-08', 'grad_norm': '0.00542664', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.02307129', 'grpo_loop_time': '420.61453247', 'ref_model_time': '114.15479279', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85929215', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.92724609', 'token_credit_video_advantage_mean': '1.327e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.501e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-08', 'kl_loss': '0.00796284', 'kl_moving': '0.00796284', 'kl_stepwise': '-2.014e-08', 'gathered_avg_reward_mean': '8.92687988', 'gathered_avg_reward_std': '1.35543668', 'gathered_video_score2_mq_reward_mean': '3.16918945', 'gathered_video_score2_mq_reward_std': '0.34554920', 'gathered_video_score2_overall_reward_mean': '8.92724609', 'gathered_video_score2_overall_reward_std': '1.35570014', 'gathered_video_score2_ta_reward_mean': '2.91174316', 'gathered_video_score2_ta_reward_std': '0.76196748', 'gathered_video_score2_vq_reward_mean': '2.84594727', 'gathered_video_score2_vq_reward_std': '0.63114423', 'gathered_ori_avg_reward_mean': '8.92687988', 'gathered_ori_avg_reward_std': '1.35543668'} +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 02:51:53.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_0/index_691_seed_10955_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 02:53:16.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:53:16.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:25.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_0/index_691_seed_10956_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:54:47.584 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 02:54:48.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.684 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 139 train_steps 1560 {'step': '139/10000', 'learning_rate': '0.00001000', 'loss': '5.215e-09', 'grad_norm': '0.00486888', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.29431152', 'grpo_loop_time': '420.38632202', 'ref_model_time': '112.86775970', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86562371', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.02636719', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.512e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.166e-09', 'kl_loss': '0.00766999', 'kl_moving': '0.00766999', 'kl_stepwise': '-1.951e-08', 'gathered_avg_reward_mean': '9.02709961', 'gathered_avg_reward_std': '1.03827095', 'gathered_video_score2_mq_reward_mean': '3.16247559', 'gathered_video_score2_mq_reward_std': '0.36925992', 'gathered_video_score2_overall_reward_mean': '9.02636719', 'gathered_video_score2_overall_reward_std': '1.03725791', 'gathered_video_score2_ta_reward_mean': '2.87329102', 'gathered_video_score2_ta_reward_std': '0.63327116', 'gathered_video_score2_vq_reward_mean': '2.99133301', 'gathered_video_score2_vq_reward_std': '0.49652055', 'gathered_ori_avg_reward_mean': '9.02709961', 'gathered_ori_avg_reward_std': '1.03827107'} +2026-06-26 03:05:40.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 03:05:40.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_0/index_7124_seed_13630_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 03:07:08.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:07:08.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:15.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_0/index_7124_seed_13631_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:08:37.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 03:08:38.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:19:28.841 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 140 train_steps 1600 {'step': '140/10000', 'learning_rate': '0.00001000', 'loss': '-1.304e-08', 'grad_norm': '0.00427179', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.72766113', 'grpo_loop_time': '418.35675049', 'ref_model_time': '113.20648193', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86739159', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03564453', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.446e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.304e-08', 'kl_loss': '0.00774111', 'kl_moving': '0.00774111', 'kl_stepwise': '-1.986e-08', 'gathered_avg_reward_mean': '9.03015137', 'gathered_avg_reward_std': '0.86822045', 'gathered_video_score2_mq_reward_mean': '3.10742188', 'gathered_video_score2_mq_reward_std': '0.29362556', 'gathered_video_score2_overall_reward_mean': '9.03564453', 'gathered_video_score2_overall_reward_std': '0.87127531', 'gathered_video_score2_ta_reward_mean': '3.02832031', 'gathered_video_score2_ta_reward_std': '0.55188054', 'gathered_video_score2_vq_reward_mean': '2.89440918', 'gathered_video_score2_vq_reward_std': '0.42383486', 'gathered_ori_avg_reward_mean': '9.03015137', 'gathered_ori_avg_reward_std': '0.86822051'} +2026-06-26 03:20:04.747 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 03:20:06.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:14.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_0/index_4029_seed_10540_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 03:21:29.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:29.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:39.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_0/index_4029_seed_10541_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:22:58.512 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 03:22:59.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.473 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 141 train_steps 1640 {'step': '141/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00594301', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.95532227', 'grpo_loop_time': '419.73391724', 'ref_model_time': '114.23197937', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85458928', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.83398438', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.423e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00730861', 'kl_moving': '0.00730861', 'kl_stepwise': '-1.934e-08', 'gathered_avg_reward_mean': '8.83703613', 'gathered_avg_reward_std': '0.93417674', 'gathered_video_score2_mq_reward_mean': '3.05346680', 'gathered_video_score2_mq_reward_std': '0.35453448', 'gathered_video_score2_overall_reward_mean': '8.83398438', 'gathered_video_score2_overall_reward_std': '0.93717241', 'gathered_video_score2_ta_reward_mean': '2.84851074', 'gathered_video_score2_ta_reward_std': '0.54647487', 'gathered_video_score2_vq_reward_mean': '2.93505859', 'gathered_video_score2_vq_reward_std': '0.33367264', 'gathered_ori_avg_reward_mean': '8.83703613', 'gathered_ori_avg_reward_std': '0.93417674'} +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 03:33:52.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_0/index_278_seed_9871_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 03:35:14.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:35:14.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:24.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_0/index_278_seed_9872_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 03:36:45.186 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:36:45.355 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 03:36:46.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.013 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 142 train_steps 1680 {'step': '142/10000', 'learning_rate': '0.00001000', 'loss': '2.235e-08', 'grad_norm': '0.00533744', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.07058716', 'grpo_loop_time': '420.61761475', 'ref_model_time': '112.44723511', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87275165', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.84814453', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.487e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.676e-08', 'kl_loss': '0.00742699', 'kl_moving': '0.00742699', 'kl_stepwise': '-1.956e-08', 'gathered_avg_reward_mean': '8.84814453', 'gathered_avg_reward_std': '1.21462226', 'gathered_video_score2_mq_reward_mean': '3.09692383', 'gathered_video_score2_mq_reward_std': '0.34951857', 'gathered_video_score2_overall_reward_mean': '8.84814453', 'gathered_video_score2_overall_reward_std': '1.21704447', 'gathered_video_score2_ta_reward_mean': '2.98815918', 'gathered_video_score2_ta_reward_std': '0.60464442', 'gathered_video_score2_vq_reward_mean': '2.76306152', 'gathered_video_score2_vq_reward_std': '0.56067628', 'gathered_ori_avg_reward_mean': '8.84814453', 'gathered_ori_avg_reward_std': '1.21462226'} +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 03:47:40.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:46.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_0/index_7277_seed_21237_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 03:49:01.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:49:01.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_0/index_7277_seed_21238_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:50:30.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 03:50:31.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.067 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 143 train_steps 1720 {'step': '143/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00705431', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.31262207', 'grpo_loop_time': '419.16345215', 'ref_model_time': '112.44873047', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85547751', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.27539062', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.473e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.00776808', 'kl_moving': '0.00776808', 'kl_stepwise': '-1.968e-08', 'gathered_avg_reward_mean': '9.27392578', 'gathered_avg_reward_std': '0.89370447', 'gathered_video_score2_mq_reward_mean': '3.17944336', 'gathered_video_score2_mq_reward_std': '0.31594768', 'gathered_video_score2_overall_reward_mean': '9.27539062', 'gathered_video_score2_overall_reward_std': '0.89218056', 'gathered_video_score2_ta_reward_mean': '3.18383789', 'gathered_video_score2_ta_reward_std': '0.44477886', 'gathered_video_score2_vq_reward_mean': '2.91064453', 'gathered_video_score2_vq_reward_std': '0.44534007', 'gathered_ori_avg_reward_mean': '9.27392578', 'gathered_ori_avg_reward_std': '0.89370447'} +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 04:01:21.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:27.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_0/index_656_seed_24741_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 04:02:43.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:43.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:03:52.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_0/index_656_seed_24742_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:04:11.680 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 04:04:12.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.604 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 144 train_steps 1760 {'step': '144/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00409629', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.85327148', 'grpo_loop_time': '422.33227539', 'ref_model_time': '113.71294403', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87048477', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.26318359', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.398e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00706523', 'kl_moving': '0.00706523', 'kl_stepwise': '-1.833e-08', 'gathered_avg_reward_mean': '9.26635742', 'gathered_avg_reward_std': '1.00375962', 'gathered_video_score2_mq_reward_mean': '3.14843750', 'gathered_video_score2_mq_reward_std': '0.30875441', 'gathered_video_score2_overall_reward_mean': '9.26318359', 'gathered_video_score2_overall_reward_std': '1.00555110', 'gathered_video_score2_ta_reward_mean': '2.98583984', 'gathered_video_score2_ta_reward_std': '0.60302377', 'gathered_video_score2_vq_reward_mean': '3.13208008', 'gathered_video_score2_vq_reward_std': '0.39138529', 'gathered_ori_avg_reward_mean': '9.26635742', 'gathered_ori_avg_reward_std': '1.00375974'} +2026-06-26 04:15:09.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 04:15:09.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:16.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_0/index_7316_seed_27300_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 04:16:32.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:32.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:17:42.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_0/index_7316_seed_27301_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:18:04.942 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 04:18:05.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.808 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 145 train_steps 1800 {'step': '145/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-10', 'grad_norm': '0.00541856', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.81027222', 'grpo_loop_time': '421.76028442', 'ref_model_time': '113.96007538', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85548234', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.02246094', 'token_credit_video_advantage_mean': '-1.339e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.416e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.260e-09', 'kl_loss': '0.00773843', 'kl_moving': '0.00773844', 'kl_stepwise': '-1.875e-08', 'gathered_avg_reward_mean': '9.02160645', 'gathered_avg_reward_std': '1.57520390', 'gathered_video_score2_mq_reward_mean': '3.07556152', 'gathered_video_score2_mq_reward_std': '0.52773255', 'gathered_video_score2_overall_reward_mean': '9.02246094', 'gathered_video_score2_overall_reward_std': '1.57538116', 'gathered_video_score2_ta_reward_mean': '3.13342285', 'gathered_video_score2_ta_reward_std': '0.73832428', 'gathered_video_score2_vq_reward_mean': '2.81262207', 'gathered_video_score2_vq_reward_std': '0.58039945', 'gathered_ori_avg_reward_mean': '9.02160645', 'gathered_ori_avg_reward_std': '1.57520390'} +2026-06-26 04:29:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 04:29:02.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:09.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_0/index_7297_seed_12289_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 04:30:27.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:27.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:33.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_0/index_7297_seed_12290_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:31:53.338 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 04:31:54.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.251 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 146 train_steps 1840 {'step': '146/10000', 'learning_rate': '0.00001000', 'loss': '1.397e-08', 'grad_norm': '0.00488194', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.38912964', 'grpo_loop_time': '421.24804688', 'ref_model_time': '112.37662506', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86706150', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.00488281', 'token_credit_video_advantage_mean': '-8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.389e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-08', 'kl_loss': '0.00794924', 'kl_moving': '0.00794924', 'kl_stepwise': '-1.869e-08', 'gathered_avg_reward_mean': '9.00305176', 'gathered_avg_reward_std': '1.36403906', 'gathered_video_score2_mq_reward_mean': '3.18847656', 'gathered_video_score2_mq_reward_std': '0.41605574', 'gathered_video_score2_overall_reward_mean': '9.00488281', 'gathered_video_score2_overall_reward_std': '1.37211406', 'gathered_video_score2_ta_reward_mean': '2.91821289', 'gathered_video_score2_ta_reward_std': '0.60828161', 'gathered_video_score2_vq_reward_mean': '2.89636230', 'gathered_video_score2_vq_reward_std': '0.58230358', 'gathered_ori_avg_reward_mean': '9.00305176', 'gathered_ori_avg_reward_std': '1.36403906'} +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 04:42:46.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:52.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_0/index_5451_seed_5374_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 04:44:05.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:44:05.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:16.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_0/index_5451_seed_5375_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 04:45:34.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:45:34.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 04:45:35.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.439 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.502 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 147 train_steps 1880 {'step': '147/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00630889', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.64425659', 'grpo_loop_time': '419.43304443', 'ref_model_time': '113.41481781', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84880584', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.30175781', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.449e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00800661', 'kl_moving': '0.00800662', 'kl_stepwise': '-1.949e-08', 'gathered_avg_reward_mean': '9.30566406', 'gathered_avg_reward_std': '1.23115647', 'gathered_video_score2_mq_reward_mean': '3.21716309', 'gathered_video_score2_mq_reward_std': '0.42646426', 'gathered_video_score2_overall_reward_mean': '9.30175781', 'gathered_video_score2_overall_reward_std': '1.23099840', 'gathered_video_score2_ta_reward_mean': '3.15612793', 'gathered_video_score2_ta_reward_std': '0.56272459', 'gathered_video_score2_vq_reward_mean': '2.93237305', 'gathered_video_score2_vq_reward_std': '0.45942718', 'gathered_ori_avg_reward_mean': '9.30566406', 'gathered_ori_avg_reward_std': '1.23115647'} +2026-06-26 04:56:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 04:56:27.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_0/index_841_seed_31574_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 04:57:52.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:52.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:58:58.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_0/index_841_seed_31575_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:59:18.363 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 04:59:19.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.379 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 148 train_steps 1920 {'step': '148/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00636339', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.38000488', 'grpo_loop_time': '418.80538940', 'ref_model_time': '112.77744293', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85452211', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.19824219', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.396e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00753819', 'kl_moving': '0.00753819', 'kl_stepwise': '-1.920e-08', 'gathered_avg_reward_mean': '9.19934082', 'gathered_avg_reward_std': '1.21980286', 'gathered_video_score2_mq_reward_mean': '3.23803711', 'gathered_video_score2_mq_reward_std': '0.38310876', 'gathered_video_score2_overall_reward_mean': '9.19824219', 'gathered_video_score2_overall_reward_std': '1.22122920', 'gathered_video_score2_ta_reward_mean': '2.85803223', 'gathered_video_score2_ta_reward_std': '0.65667504', 'gathered_video_score2_vq_reward_mean': '3.10327148', 'gathered_video_score2_vq_reward_std': '0.57192636', 'gathered_ori_avg_reward_mean': '9.19934082', 'gathered_ori_avg_reward_std': '1.21980274'} +2026-06-26 05:10:10.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 05:10:10.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:16.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_0/index_2883_seed_31728_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 05:11:30.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:30.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:12:42.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_0/index_2883_seed_31729_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 05:13:02.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:13:02.825 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 05:13:04.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.830 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 149 train_steps 1960 {'step': '149/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00682810', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.22998047', 'grpo_loop_time': '423.26220703', 'ref_model_time': '113.89624786', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86394292', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.43920898', 'token_credit_video_advantage_mean': '9.313e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.558e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00819735', 'kl_moving': '0.00819735', 'kl_stepwise': '-2.017e-08', 'gathered_avg_reward_mean': '9.43725586', 'gathered_avg_reward_std': '1.63099623', 'gathered_video_score2_mq_reward_mean': '3.17590332', 'gathered_video_score2_mq_reward_std': '0.54972404', 'gathered_video_score2_overall_reward_mean': '9.43920898', 'gathered_video_score2_overall_reward_std': '1.63189209', 'gathered_video_score2_ta_reward_mean': '3.23718262', 'gathered_video_score2_ta_reward_std': '0.71496034', 'gathered_video_score2_vq_reward_mean': '3.02416992', 'gathered_video_score2_vq_reward_std': '0.61053938', 'gathered_ori_avg_reward_mean': '9.43725586', 'gathered_ori_avg_reward_std': '1.63099623'} +2026-06-26 05:24:02.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 05:24:02.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:09.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_0/index_9953_seed_18575_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 05:25:29.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:29.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:35.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_0/index_9953_seed_18576_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:26:58.104 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 05:26:59.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:37:50.203 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 150 train_steps 2000 {'step': '150/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00504411', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.69592285', 'grpo_loop_time': '418.89797974', 'ref_model_time': '113.88749695', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86542535', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.50048828', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.425e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00846306', 'kl_moving': '0.00846306', 'kl_stepwise': '-1.963e-08', 'gathered_avg_reward_mean': '8.50170898', 'gathered_avg_reward_std': '1.00477076', 'gathered_video_score2_mq_reward_mean': '2.96752930', 'gathered_video_score2_mq_reward_std': '0.30491659', 'gathered_video_score2_overall_reward_mean': '8.50048828', 'gathered_video_score2_overall_reward_std': '1.00504541', 'gathered_video_score2_ta_reward_mean': '2.80297852', 'gathered_video_score2_ta_reward_std': '0.68345904', 'gathered_video_score2_vq_reward_mean': '2.73120117', 'gathered_video_score2_vq_reward_std': '0.45574138', 'gathered_ori_avg_reward_mean': '8.50170898', 'gathered_ori_avg_reward_std': '1.00477087'} +2026-06-26 05:38:47.492 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 150 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-150 +2026-06-26 05:38:47.642 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 150: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-150 +2026-06-26 05:38:50.482 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-125 +2026-06-26 05:39:26.527 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 05:39:29.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:36.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_0/index_8040_seed_7961_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 05:40:55.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:55.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:00.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_0/index_8040_seed_7962_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:42:19.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 05:42:20.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.414 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 151 train_steps 2040 {'step': '151/10000', 'learning_rate': '0.00001000', 'loss': '-1.024e-08', 'grad_norm': '0.01100376', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.26016235', 'grpo_loop_time': '419.14199829', 'ref_model_time': '112.09495544', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87285084', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.06005859', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.467e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.024e-08', 'kl_loss': '0.00767556', 'kl_moving': '0.00767556', 'kl_stepwise': '-1.928e-08', 'gathered_avg_reward_mean': '9.06774902', 'gathered_avg_reward_std': '1.04908907', 'gathered_video_score2_mq_reward_mean': '3.06335449', 'gathered_video_score2_mq_reward_std': '0.40247199', 'gathered_video_score2_overall_reward_mean': '9.06005859', 'gathered_video_score2_overall_reward_std': '1.04626703', 'gathered_video_score2_ta_reward_mean': '3.16577148', 'gathered_video_score2_ta_reward_std': '0.56322426', 'gathered_video_score2_vq_reward_mean': '2.83862305', 'gathered_video_score2_vq_reward_std': '0.33001518', 'gathered_ori_avg_reward_mean': '9.06774902', 'gathered_ori_avg_reward_std': '1.04908907'} +2026-06-26 05:53:10.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 05:53:10.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:17.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_0/index_7704_seed_32713_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 05:54:32.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:32.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:39.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_0/index_7704_seed_32714_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:55:59.973 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 05:56:00.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.223 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 152 train_steps 2080 {'step': '152/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00526060', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.03628540', 'grpo_loop_time': '419.91992188', 'ref_model_time': '112.79663849', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85605061', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.90917969', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.415e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00689378', 'kl_moving': '0.00689378', 'kl_stepwise': '-1.987e-08', 'gathered_avg_reward_mean': '8.90576172', 'gathered_avg_reward_std': '0.98556191', 'gathered_video_score2_mq_reward_mean': '3.13012695', 'gathered_video_score2_mq_reward_std': '0.31234193', 'gathered_video_score2_overall_reward_mean': '8.90917969', 'gathered_video_score2_overall_reward_std': '0.98639917', 'gathered_video_score2_ta_reward_mean': '2.79528809', 'gathered_video_score2_ta_reward_std': '0.61688489', 'gathered_video_score2_vq_reward_mean': '2.98034668', 'gathered_video_score2_vq_reward_std': '0.51597261', 'gathered_ori_avg_reward_mean': '8.90576172', 'gathered_ori_avg_reward_std': '0.98556191'} +2026-06-26 06:06:52.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 06:06:52.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:58.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_0/index_5842_seed_17139_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 06:08:12.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:08:12.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:22.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_0/index_5842_seed_17140_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:09:44.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 06:09:44.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.798 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 153 train_steps 2120 {'step': '153/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00557496', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.39572144', 'grpo_loop_time': '420.99780273', 'ref_model_time': '113.00323486', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86812055', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.84472656', 'token_credit_video_advantage_mean': '1.304e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.513e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00739579', 'kl_moving': '0.00739579', 'kl_stepwise': '-1.900e-08', 'gathered_avg_reward_mean': '8.84155273', 'gathered_avg_reward_std': '1.02672195', 'gathered_video_score2_mq_reward_mean': '3.12097168', 'gathered_video_score2_mq_reward_std': '0.38935339', 'gathered_video_score2_overall_reward_mean': '8.84472656', 'gathered_video_score2_overall_reward_std': '1.02749920', 'gathered_video_score2_ta_reward_mean': '2.95043945', 'gathered_video_score2_ta_reward_std': '0.46262729', 'gathered_video_score2_vq_reward_mean': '2.77014160', 'gathered_video_score2_vq_reward_std': '0.45728534', 'gathered_ori_avg_reward_mean': '8.84155273', 'gathered_ori_avg_reward_std': '1.02672207'} +2026-06-26 06:20:39.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 06:20:39.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:39.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_0/index_608_seed_28453_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 06:22:00.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:22:00.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:10.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_0/index_608_seed_28454_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 06:23:30.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:23:30.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 06:23:31.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.693 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 154 train_steps 2160 {'step': '154/10000', 'learning_rate': '0.00001000', 'loss': '8.848e-09', 'grad_norm': '0.00562955', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.58215332', 'grpo_loop_time': '419.31152344', 'ref_model_time': '112.26452637', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86641037', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14599609', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.504e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.985e-09', 'kl_loss': '0.00762180', 'kl_moving': '0.00762180', 'kl_stepwise': '-1.926e-08', 'gathered_avg_reward_mean': '9.14477539', 'gathered_avg_reward_std': '0.84826350', 'gathered_video_score2_mq_reward_mean': '3.15039062', 'gathered_video_score2_mq_reward_std': '0.25869140', 'gathered_video_score2_overall_reward_mean': '9.14599609', 'gathered_video_score2_overall_reward_std': '0.84555751', 'gathered_video_score2_ta_reward_mean': '3.04541016', 'gathered_video_score2_ta_reward_std': '0.51594096', 'gathered_video_score2_vq_reward_mean': '2.94897461', 'gathered_video_score2_vq_reward_std': '0.46225142', 'gathered_ori_avg_reward_mean': '9.14477539', 'gathered_ori_avg_reward_std': '0.84826356'} +2026-06-26 06:34:21.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 06:34:21.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_0/index_731_seed_19095_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 06:35:43.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:43.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:36:51.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_0/index_731_seed_19096_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 06:37:10.499 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:37:10.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 06:37:11.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:09.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.560 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 155 train_steps 2200 {'step': '155/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00703498', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.96243286', 'grpo_loop_time': '422.49166870', 'ref_model_time': '114.47926331', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86451340', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.19042969', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.399e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00841304', 'kl_moving': '0.00841304', 'kl_stepwise': '-1.873e-08', 'gathered_avg_reward_mean': '8.19287109', 'gathered_avg_reward_std': '1.45188594', 'gathered_video_score2_mq_reward_mean': '2.88415527', 'gathered_video_score2_mq_reward_std': '0.49162051', 'gathered_video_score2_overall_reward_mean': '8.19042969', 'gathered_video_score2_overall_reward_std': '1.45259690', 'gathered_video_score2_ta_reward_mean': '2.76831055', 'gathered_video_score2_ta_reward_std': '0.67116809', 'gathered_video_score2_vq_reward_mean': '2.54040527', 'gathered_video_score2_vq_reward_std': '0.60214174', 'gathered_ori_avg_reward_mean': '8.19287109', 'gathered_ori_avg_reward_std': '1.45188594'} +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 06:48:09.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:16.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_0/index_4724_seed_8573_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 06:49:29.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:29.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:50:44.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_0/index_4724_seed_8574_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:51:02.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 06:51:03.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:02.964 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 156 train_steps 2240 {'step': '156/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00607558', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.08099365', 'grpo_loop_time': '423.63079834', 'ref_model_time': '114.19063568', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85794514', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.18017578', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.378e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.00806994', 'kl_moving': '0.00806994', 'kl_stepwise': '-1.893e-08', 'gathered_avg_reward_mean': '9.18640137', 'gathered_avg_reward_std': '0.95951414', 'gathered_video_score2_mq_reward_mean': '3.17492676', 'gathered_video_score2_mq_reward_std': '0.30986321', 'gathered_video_score2_overall_reward_mean': '9.18017578', 'gathered_video_score2_overall_reward_std': '0.96194911', 'gathered_video_score2_ta_reward_mean': '2.98315430', 'gathered_video_score2_ta_reward_std': '0.64379132', 'gathered_video_score2_vq_reward_mean': '3.02832031', 'gathered_video_score2_vq_reward_std': '0.33601555', 'gathered_ori_avg_reward_mean': '9.18640137', 'gathered_ori_avg_reward_std': '0.95951414'} +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 07:02:03.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:09.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_0/index_8869_seed_19024_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 07:03:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:23.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:34.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_0/index_8869_seed_19025_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 07:04:54.348 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:04:54.520 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 07:04:55.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.401 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 157 train_steps 2280 {'step': '157/10000', 'learning_rate': '0.00001000', 'loss': '6.519e-09', 'grad_norm': '0.00519063', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.58428955', 'grpo_loop_time': '419.96084595', 'ref_model_time': '114.49771881', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87931979', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01074219', 'token_credit_video_advantage_mean': '3.376e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.456e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00804466', 'kl_moving': '0.00804466', 'kl_stepwise': '-1.894e-08', 'gathered_avg_reward_mean': '9.01098633', 'gathered_avg_reward_std': '0.88429093', 'gathered_video_score2_mq_reward_mean': '3.13549805', 'gathered_video_score2_mq_reward_std': '0.24272330', 'gathered_video_score2_overall_reward_mean': '9.01074219', 'gathered_video_score2_overall_reward_std': '0.88404262', 'gathered_video_score2_ta_reward_mean': '2.95129395', 'gathered_video_score2_ta_reward_std': '0.60729742', 'gathered_video_score2_vq_reward_mean': '2.92419434', 'gathered_video_score2_vq_reward_std': '0.42187428', 'gathered_ori_avg_reward_mean': '9.01098633', 'gathered_ori_avg_reward_std': '0.88429093'} +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 07:15:48.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:54.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_0/index_7324_seed_9609_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 07:17:11.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:17:11.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:21.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_0/index_7324_seed_9610_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:18:40.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 07:18:41.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.001 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 158 train_steps 2320 {'step': '158/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00443569', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.67807007', 'grpo_loop_time': '418.91479492', 'ref_model_time': '112.57518005', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86349487', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.25683594', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.418e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00724745', 'kl_moving': '0.00724745', 'kl_stepwise': '-1.878e-08', 'gathered_avg_reward_mean': '9.25817871', 'gathered_avg_reward_std': '1.04422653', 'gathered_video_score2_mq_reward_mean': '3.22509766', 'gathered_video_score2_mq_reward_std': '0.35853246', 'gathered_video_score2_overall_reward_mean': '9.25683594', 'gathered_video_score2_overall_reward_std': '1.04254389', 'gathered_video_score2_ta_reward_mean': '2.95190430', 'gathered_video_score2_ta_reward_std': '0.57604975', 'gathered_video_score2_vq_reward_mean': '3.08117676', 'gathered_video_score2_vq_reward_std': '0.48473245', 'gathered_ori_avg_reward_mean': '9.25817871', 'gathered_ori_avg_reward_std': '1.04422653'} +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 07:29:32.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:38.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_0/index_4989_seed_16546_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 07:30:53.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:53.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:05.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_0/index_4989_seed_16547_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:32:30.267 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 07:32:31.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.440 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 159 train_steps 2360 {'step': '159/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00655071', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.33721924', 'grpo_loop_time': '419.00921631', 'ref_model_time': '112.56586456', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87921995', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.47314453', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.383e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00841215', 'kl_moving': '0.00841215', 'kl_stepwise': '-1.916e-08', 'gathered_avg_reward_mean': '8.47106934', 'gathered_avg_reward_std': '1.09206998', 'gathered_video_score2_mq_reward_mean': '3.04846191', 'gathered_video_score2_mq_reward_std': '0.36368841', 'gathered_video_score2_overall_reward_mean': '8.47314453', 'gathered_video_score2_overall_reward_std': '1.09466851', 'gathered_video_score2_ta_reward_mean': '2.71154785', 'gathered_video_score2_ta_reward_std': '0.52874583', 'gathered_video_score2_vq_reward_mean': '2.71105957', 'gathered_video_score2_vq_reward_std': '0.49652126', 'gathered_ori_avg_reward_mean': '8.47106934', 'gathered_ori_avg_reward_std': '1.09206998'} +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 07:43:21.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:27.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_0/index_4108_seed_23778_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 07:44:42.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:42.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:45:53.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_0/index_4108_seed_23779_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:46:14.696 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 07:46:15.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:08.452 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 160 train_steps 2400 {'step': '160/10000', 'learning_rate': '0.00001000', 'loss': '1.024e-08', 'grad_norm': '0.00650943', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.39514160', 'grpo_loop_time': '420.54281616', 'ref_model_time': '113.65727234', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86971873', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.81542969', 'token_credit_video_advantage_mean': '2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.495e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00832022', 'kl_moving': '0.00832022', 'kl_stepwise': '-2.008e-08', 'gathered_avg_reward_mean': '8.81250000', 'gathered_avg_reward_std': '1.24285376', 'gathered_video_score2_mq_reward_mean': '3.03015137', 'gathered_video_score2_mq_reward_std': '0.42236722', 'gathered_video_score2_overall_reward_mean': '8.81542969', 'gathered_video_score2_overall_reward_std': '1.24494481', 'gathered_video_score2_ta_reward_mean': '3.05895996', 'gathered_video_score2_ta_reward_std': '0.60021651', 'gathered_video_score2_vq_reward_mean': '2.72338867', 'gathered_video_score2_vq_reward_std': '0.56995404', 'gathered_ori_avg_reward_mean': '8.81250000', 'gathered_ori_avg_reward_std': '1.24285376'} +2026-06-26 07:57:43.676 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 07:57:45.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:45.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:53.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_0/index_7288_seed_20155_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 07:59:07.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:59:07.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:19.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_0/index_7288_seed_20156_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:00:39.683 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 08:00:40.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:32.881 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 161 train_steps 2440 {'step': '161/10000', 'learning_rate': '0.00001000', 'loss': '1.490e-08', 'grad_norm': '0.01333657', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.70892334', 'grpo_loop_time': '420.07742310', 'ref_model_time': '112.98210907', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87488657', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.19140625', 'token_credit_video_advantage_mean': '-5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.409e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.490e-08', 'kl_loss': '0.00799829', 'kl_moving': '0.00799829', 'kl_stepwise': '-1.939e-08', 'gathered_avg_reward_mean': '9.18957520', 'gathered_avg_reward_std': '1.09416783', 'gathered_video_score2_mq_reward_mean': '3.16918945', 'gathered_video_score2_mq_reward_std': '0.31096491', 'gathered_video_score2_overall_reward_mean': '9.19140625', 'gathered_video_score2_overall_reward_std': '1.09118700', 'gathered_video_score2_ta_reward_mean': '3.18249512', 'gathered_video_score2_ta_reward_std': '0.51074755', 'gathered_video_score2_vq_reward_mean': '2.83789062', 'gathered_video_score2_vq_reward_std': '0.54426867', 'gathered_ori_avg_reward_mean': '9.18957520', 'gathered_ori_avg_reward_std': '1.09416783'} +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 08:11:33.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:39.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_0/index_7919_seed_4851_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 08:12:55.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:55.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:03.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_0/index_7919_seed_4852_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:14:25.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 08:14:26.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.070 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 162 train_steps 2480 {'step': '162/10000', 'learning_rate': '0.00001000', 'loss': '-6.519e-09', 'grad_norm': '0.00508187', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.49362183', 'grpo_loop_time': '418.87554932', 'ref_model_time': '112.41497803', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86922568', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.81982422', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.499e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.00787032', 'kl_moving': '0.00787032', 'kl_stepwise': '-1.970e-08', 'gathered_avg_reward_mean': '8.82165527', 'gathered_avg_reward_std': '1.18089747', 'gathered_video_score2_mq_reward_mean': '3.02233887', 'gathered_video_score2_mq_reward_std': '0.38305375', 'gathered_video_score2_overall_reward_mean': '8.81982422', 'gathered_video_score2_overall_reward_std': '1.18345511', 'gathered_video_score2_ta_reward_mean': '3.00854492', 'gathered_video_score2_ta_reward_std': '0.63025546', 'gathered_video_score2_vq_reward_mean': '2.79077148', 'gathered_video_score2_vq_reward_std': '0.51122922', 'gathered_ori_avg_reward_mean': '8.82165527', 'gathered_ori_avg_reward_std': '1.18089747'} +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 08:25:18.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:24.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_0/index_2641_seed_15967_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 08:26:40.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:40.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:27:48.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_0/index_2641_seed_15968_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:28:07.189 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 08:28:08.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.656 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.267 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 163 train_steps 2520 {'step': '163/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00442420', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.13394165', 'grpo_loop_time': '421.92184448', 'ref_model_time': '112.61677551', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87492990', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14990234', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00760136', 'kl_moving': '0.00760136', 'kl_stepwise': '-1.793e-08', 'gathered_avg_reward_mean': '9.14953613', 'gathered_avg_reward_std': '1.24794900', 'gathered_video_score2_mq_reward_mean': '3.25195312', 'gathered_video_score2_mq_reward_std': '0.42669860', 'gathered_video_score2_overall_reward_mean': '9.14990234', 'gathered_video_score2_overall_reward_std': '1.24931836', 'gathered_video_score2_ta_reward_mean': '2.93139648', 'gathered_video_score2_ta_reward_std': '0.66522682', 'gathered_video_score2_vq_reward_mean': '2.96618652', 'gathered_video_score2_vq_reward_std': '0.52584612', 'gathered_ori_avg_reward_mean': '9.14953613', 'gathered_ori_avg_reward_std': '1.24794900'} +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 08:39:03.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:09.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_0/index_2475_seed_8902_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 08:40:27.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:27.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:36.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_0/index_2475_seed_8903_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:41:56.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 08:41:57.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.821 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 164 train_steps 2560 {'step': '164/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00452275', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.27734375', 'grpo_loop_time': '419.34179688', 'ref_model_time': '114.64203644', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87140477', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.82226562', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.423e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.304e-08', 'kl_loss': '0.00746708', 'kl_moving': '0.00746708', 'kl_stepwise': '-1.973e-08', 'gathered_avg_reward_mean': '8.81994629', 'gathered_avg_reward_std': '1.36763215', 'gathered_video_score2_mq_reward_mean': '3.02709961', 'gathered_video_score2_mq_reward_std': '0.51407641', 'gathered_video_score2_overall_reward_mean': '8.82226562', 'gathered_video_score2_overall_reward_std': '1.36440420', 'gathered_video_score2_ta_reward_mean': '2.98144531', 'gathered_video_score2_ta_reward_std': '0.64087433', 'gathered_video_score2_vq_reward_mean': '2.81140137', 'gathered_video_score2_vq_reward_std': '0.54906768', 'gathered_ori_avg_reward_mean': '8.81994629', 'gathered_ori_avg_reward_std': '1.36763203'} +2026-06-26 08:52:49.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 08:52:49.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_0/index_5671_seed_5319_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 08:54:15.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:54:15.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:22.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_0/index_5671_seed_5320_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 08:55:44.234 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:55:44.392 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 08:55:45.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.126 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 165 train_steps 2600 {'step': '165/10000', 'learning_rate': '0.00001000', 'loss': '6.985e-10', 'grad_norm': '0.00623586', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.84875488', 'grpo_loop_time': '420.45980835', 'ref_model_time': '112.65375519', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85404366', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.37792969', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.439e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.985e-10', 'kl_loss': '0.00744969', 'kl_moving': '0.00744969', 'kl_stepwise': '-1.927e-08', 'gathered_avg_reward_mean': '9.37609863', 'gathered_avg_reward_std': '1.32193017', 'gathered_video_score2_mq_reward_mean': '3.24560547', 'gathered_video_score2_mq_reward_std': '0.45262355', 'gathered_video_score2_overall_reward_mean': '9.37792969', 'gathered_video_score2_overall_reward_std': '1.32198393', 'gathered_video_score2_ta_reward_mean': '3.17553711', 'gathered_video_score2_ta_reward_std': '0.56459498', 'gathered_video_score2_vq_reward_mean': '2.95495605', 'gathered_video_score2_vq_reward_std': '0.65092230', 'gathered_ori_avg_reward_mean': '9.37609863', 'gathered_ori_avg_reward_std': '1.32193017'} +2026-06-26 09:06:39.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 09:06:39.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_0/index_9828_seed_19026_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 09:08:00.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:08:00.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:11.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_0/index_9828_seed_19027_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:09:32.611 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 09:09:33.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.631 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 166 train_steps 2640 {'step': '166/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00556942', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.04251099', 'grpo_loop_time': '418.33285522', 'ref_model_time': '112.13878632', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87850273', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.94946289', 'token_credit_video_advantage_mean': '-1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.449e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00778202', 'kl_moving': '0.00778202', 'kl_stepwise': '-1.995e-08', 'gathered_avg_reward_mean': '8.95141602', 'gathered_avg_reward_std': '1.15252209', 'gathered_video_score2_mq_reward_mean': '3.08325195', 'gathered_video_score2_mq_reward_std': '0.32515928', 'gathered_video_score2_overall_reward_mean': '8.94946289', 'gathered_video_score2_overall_reward_std': '1.15334225', 'gathered_video_score2_ta_reward_mean': '3.11303711', 'gathered_video_score2_ta_reward_std': '0.63330507', 'gathered_video_score2_vq_reward_mean': '2.75512695', 'gathered_video_score2_vq_reward_std': '0.46382344', 'gathered_ori_avg_reward_mean': '8.95141602', 'gathered_ori_avg_reward_std': '1.15252209'} +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 09:20:22.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:28.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_0/index_8959_seed_28930_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 09:21:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:43.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:22:52.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_0/index_8959_seed_28931_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:23:11.996 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 09:23:12.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.194 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 167 train_steps 2680 {'step': '167/10000', 'learning_rate': '0.00001000', 'loss': '1.164e-09', 'grad_norm': '0.00443634', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.96582031', 'grpo_loop_time': '421.42849731', 'ref_model_time': '114.04237366', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84798092', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.93212891', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.421e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.985e-10', 'kl_loss': '0.00752321', 'kl_moving': '0.00752321', 'kl_stepwise': '-1.869e-08', 'gathered_avg_reward_mean': '8.93188477', 'gathered_avg_reward_std': '1.39416432', 'gathered_video_score2_mq_reward_mean': '3.10253906', 'gathered_video_score2_mq_reward_std': '0.41524810', 'gathered_video_score2_overall_reward_mean': '8.93212891', 'gathered_video_score2_overall_reward_std': '1.39105892', 'gathered_video_score2_ta_reward_mean': '3.04614258', 'gathered_video_score2_ta_reward_std': '0.64501303', 'gathered_video_score2_vq_reward_mean': '2.78320312', 'gathered_video_score2_vq_reward_std': '0.59395236', 'gathered_ori_avg_reward_mean': '8.93188477', 'gathered_ori_avg_reward_std': '1.39416432'} +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 09:34:09.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:16.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_0/index_4889_seed_4284_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 09:35:30.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:30.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:36:42.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_0/index_4889_seed_4285_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 09:37:03.159 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:37:03.297 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 09:37:04.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.621 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 168 train_steps 2720 {'step': '168/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00583665', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.64614868', 'grpo_loop_time': '419.95629883', 'ref_model_time': '112.84716797', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88003802', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01513672', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.453e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.794e-09', 'kl_loss': '0.00814684', 'kl_moving': '0.00814685', 'kl_stepwise': '-1.937e-08', 'gathered_avg_reward_mean': '9.01232910', 'gathered_avg_reward_std': '1.08763874', 'gathered_video_score2_mq_reward_mean': '3.17871094', 'gathered_video_score2_mq_reward_std': '0.33108637', 'gathered_video_score2_overall_reward_mean': '9.01513672', 'gathered_video_score2_overall_reward_std': '1.08924842', 'gathered_video_score2_ta_reward_mean': '2.99072266', 'gathered_video_score2_ta_reward_std': '0.62588418', 'gathered_video_score2_vq_reward_mean': '2.84289551', 'gathered_video_score2_vq_reward_std': '0.53120893', 'gathered_ori_avg_reward_mean': '9.01232910', 'gathered_ori_avg_reward_std': '1.08763885'} +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 09:47:56.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_0/index_1455_seed_8802_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 09:49:23.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:23.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:29.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_0/index_1455_seed_8803_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 09:50:50.361 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:50:50.489 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 09:50:51.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.022 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 169 train_steps 2760 {'step': '169/10000', 'learning_rate': '0.00001000', 'loss': '1.304e-08', 'grad_norm': '0.00487394', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.90676880', 'grpo_loop_time': '420.81710815', 'ref_model_time': '112.54245758', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85674012', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.58886719', 'token_credit_video_advantage_mean': '-7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.473e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.304e-08', 'kl_loss': '0.00742072', 'kl_moving': '0.00742073', 'kl_stepwise': '-1.918e-08', 'gathered_avg_reward_mean': '8.59265137', 'gathered_avg_reward_std': '1.73250818', 'gathered_video_score2_mq_reward_mean': '3.06506348', 'gathered_video_score2_mq_reward_std': '0.61672384', 'gathered_video_score2_overall_reward_mean': '8.58886719', 'gathered_video_score2_overall_reward_std': '1.73230815', 'gathered_video_score2_ta_reward_mean': '2.69055176', 'gathered_video_score2_ta_reward_std': '0.68021995', 'gathered_video_score2_vq_reward_mean': '2.83703613', 'gathered_video_score2_vq_reward_std': '0.70336348', 'gathered_ori_avg_reward_mean': '8.59265137', 'gathered_ori_avg_reward_std': '1.73250818'} +2026-06-26 10:01:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 10:01:45.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:51.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_0/index_3653_seed_9722_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 10:03:06.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:03:06.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:15.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_0/index_3653_seed_9723_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:04:34.193 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 10:04:35.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:24.612 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 170 train_steps 2800 {'step': '170/10000', 'learning_rate': '0.00001000', 'loss': '1.583e-08', 'grad_norm': '0.00495629', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.56158447', 'grpo_loop_time': '419.10916138', 'ref_model_time': '111.98095703', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87736458', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.81445312', 'token_credit_video_advantage_mean': '-8.848e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.448e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.583e-08', 'kl_loss': '0.00783056', 'kl_moving': '0.00783057', 'kl_stepwise': '-1.969e-08', 'gathered_avg_reward_mean': '8.81567383', 'gathered_avg_reward_std': '0.77296603', 'gathered_video_score2_mq_reward_mean': '3.14221191', 'gathered_video_score2_mq_reward_std': '0.34045827', 'gathered_video_score2_overall_reward_mean': '8.81445312', 'gathered_video_score2_overall_reward_std': '0.77373093', 'gathered_video_score2_ta_reward_mean': '2.81518555', 'gathered_video_score2_ta_reward_std': '0.62611479', 'gathered_video_score2_vq_reward_mean': '2.85827637', 'gathered_video_score2_vq_reward_std': '0.58063990', 'gathered_ori_avg_reward_mean': '8.81567383', 'gathered_ori_avg_reward_std': '0.77296597'} +2026-06-26 10:16:00.108 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 10:16:01.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:09.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_0/index_5473_seed_18408_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 10:17:23.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:23.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:35.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_0/index_5473_seed_18409_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:18:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 10:18:59.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.342 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 171 train_steps 2840 {'step': '171/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00588703', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.48046875', 'grpo_loop_time': '419.95382690', 'ref_model_time': '114.36861420', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85367239', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.54492188', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.417e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00762365', 'kl_moving': '0.00762365', 'kl_stepwise': '-1.858e-08', 'gathered_avg_reward_mean': '9.54626465', 'gathered_avg_reward_std': '1.10044622', 'gathered_video_score2_mq_reward_mean': '3.28027344', 'gathered_video_score2_mq_reward_std': '0.36917549', 'gathered_video_score2_overall_reward_mean': '9.54492188', 'gathered_video_score2_overall_reward_std': '1.09761977', 'gathered_video_score2_ta_reward_mean': '3.27026367', 'gathered_video_score2_ta_reward_std': '0.45766249', 'gathered_video_score2_vq_reward_mean': '2.99572754', 'gathered_video_score2_vq_reward_std': '0.52834457', 'gathered_ori_avg_reward_mean': '9.54626465', 'gathered_ori_avg_reward_std': '1.10044622'} +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 10:29:52.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:58.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_0/index_5902_seed_4454_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 10:31:13.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:31:13.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:23.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_0/index_5902_seed_4455_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:32:43.485 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 10:32:44.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.406 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 172 train_steps 2880 {'step': '172/10000', 'learning_rate': '0.00001000', 'loss': '1.583e-08', 'grad_norm': '0.00482409', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.34149170', 'grpo_loop_time': '419.64471436', 'ref_model_time': '112.80657959', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87473798', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.96777344', 'token_credit_video_advantage_mean': '1.164e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.417e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.583e-08', 'kl_loss': '0.00751633', 'kl_moving': '0.00751634', 'kl_stepwise': '-1.962e-08', 'gathered_avg_reward_mean': '8.96606445', 'gathered_avg_reward_std': '1.20644104', 'gathered_video_score2_mq_reward_mean': '3.06457520', 'gathered_video_score2_mq_reward_std': '0.44589943', 'gathered_video_score2_overall_reward_mean': '8.96777344', 'gathered_video_score2_overall_reward_std': '1.20739114', 'gathered_video_score2_ta_reward_mean': '3.04052734', 'gathered_video_score2_ta_reward_std': '0.61065990', 'gathered_video_score2_vq_reward_mean': '2.86096191', 'gathered_video_score2_vq_reward_std': '0.48628533', 'gathered_ori_avg_reward_mean': '8.96606445', 'gathered_ori_avg_reward_std': '1.20644104'} +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 10:43:37.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:43.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_0/index_9363_seed_11532_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 10:45:03.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:45:03.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:08.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_0/index_9363_seed_11533_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 10:46:28.281 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:46:28.450 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 10:46:29.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.152 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 173 train_steps 2920 {'step': '173/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00823280', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.31185913', 'grpo_loop_time': '419.98889160', 'ref_model_time': '112.77871704', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86647797', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03857422', 'token_credit_video_advantage_mean': '1.304e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.423e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00734739', 'kl_moving': '0.00734739', 'kl_stepwise': '-1.976e-08', 'gathered_avg_reward_mean': '9.03369141', 'gathered_avg_reward_std': '1.36731291', 'gathered_video_score2_mq_reward_mean': '3.10583496', 'gathered_video_score2_mq_reward_std': '0.44876832', 'gathered_video_score2_overall_reward_mean': '9.03857422', 'gathered_video_score2_overall_reward_std': '1.37082934', 'gathered_video_score2_ta_reward_mean': '3.02429199', 'gathered_video_score2_ta_reward_std': '0.68055147', 'gathered_video_score2_vq_reward_mean': '2.90356445', 'gathered_video_score2_vq_reward_std': '0.60463059', 'gathered_ori_avg_reward_mean': '9.03369141', 'gathered_ori_avg_reward_std': '1.36731291'} +2026-06-26 10:57:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 10:57:21.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_0/index_4474_seed_30408_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 10:58:43.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:43.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:59:56.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_0/index_4474_seed_30409_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:00:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 11:00:18.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.701 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 174 train_steps 2960 {'step': '174/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00671278', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.23034668', 'grpo_loop_time': '419.16064453', 'ref_model_time': '112.38197327', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87198460', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.00830078', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.499e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.490e-08', 'kl_loss': '0.00710468', 'kl_moving': '0.00710468', 'kl_stepwise': '-2.043e-08', 'gathered_avg_reward_mean': '9.00427246', 'gathered_avg_reward_std': '1.36305249', 'gathered_video_score2_mq_reward_mean': '3.11718750', 'gathered_video_score2_mq_reward_std': '0.44283637', 'gathered_video_score2_overall_reward_mean': '9.00830078', 'gathered_video_score2_overall_reward_std': '1.36271858', 'gathered_video_score2_ta_reward_mean': '2.91345215', 'gathered_video_score2_ta_reward_std': '0.59062469', 'gathered_video_score2_vq_reward_mean': '2.97363281', 'gathered_video_score2_vq_reward_std': '0.58458424', 'gathered_ori_avg_reward_mean': '9.00427246', 'gathered_ori_avg_reward_std': '1.36305249'} +2026-06-26 11:11:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 11:11:08.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:15.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_0/index_2055_seed_18410_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 11:12:33.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:33.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:39.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_0/index_2055_seed_18411_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:13:58.598 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 11:13:59.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:24:50.950 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 175 train_steps 3000 {'step': '175/10000', 'learning_rate': '0.00001000', 'loss': '4.843e-09', 'grad_norm': '0.00473121', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.43298340', 'grpo_loop_time': '419.69555664', 'ref_model_time': '112.99590302', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88310450', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.26318359', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.436e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.843e-09', 'kl_loss': '0.00719631', 'kl_moving': '0.00719631', 'kl_stepwise': '-1.953e-08', 'gathered_avg_reward_mean': '9.26257324', 'gathered_avg_reward_std': '1.00215745', 'gathered_video_score2_mq_reward_mean': '3.16674805', 'gathered_video_score2_mq_reward_std': '0.24269187', 'gathered_video_score2_overall_reward_mean': '9.26318359', 'gathered_video_score2_overall_reward_std': '1.00495911', 'gathered_video_score2_ta_reward_mean': '3.09716797', 'gathered_video_score2_ta_reward_std': '0.58274007', 'gathered_video_score2_vq_reward_mean': '2.99865723', 'gathered_video_score2_vq_reward_std': '0.44838908', 'gathered_ori_avg_reward_mean': '9.26257324', 'gathered_ori_avg_reward_std': '1.00215733'} +2026-06-26 11:26:05.046 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 175 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-175 +2026-06-26 11:26:05.178 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 175: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-175 +2026-06-26 11:26:08.007 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-150 +2026-06-26 11:26:08.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 11:26:08.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:15.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_0/index_2964_seed_16199_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 11:27:30.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:30.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:28:41.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_0/index_2964_seed_16200_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:29:02.229 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 11:29:03.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.807 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 176 train_steps 3040 {'step': '176/10000', 'learning_rate': '0.00001000', 'loss': '6.752e-09', 'grad_norm': '0.00574597', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.80627441', 'grpo_loop_time': '420.14288330', 'ref_model_time': '112.61477661', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84980214', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.98095703', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.396e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00786322', 'kl_moving': '0.00786322', 'kl_stepwise': '-2.002e-08', 'gathered_avg_reward_mean': '8.97644043', 'gathered_avg_reward_std': '0.87157476', 'gathered_video_score2_mq_reward_mean': '3.10644531', 'gathered_video_score2_mq_reward_std': '0.27144748', 'gathered_video_score2_overall_reward_mean': '8.98095703', 'gathered_video_score2_overall_reward_std': '0.87476659', 'gathered_video_score2_ta_reward_mean': '3.03784180', 'gathered_video_score2_ta_reward_std': '0.62201315', 'gathered_video_score2_vq_reward_mean': '2.83215332', 'gathered_video_score2_vq_reward_std': '0.44719738', 'gathered_ori_avg_reward_mean': '8.97644043', 'gathered_ori_avg_reward_std': '0.87157476'} +2026-06-26 11:39:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 11:39:55.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:01.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_0/index_6685_seed_4210_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 11:41:15.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:15.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:25.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_0/index_6685_seed_4211_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:42:43.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 11:42:44.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.748 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 177 train_steps 3080 {'step': '177/10000', 'learning_rate': '0.00001000', 'loss': '-8.382e-09', 'grad_norm': '0.00590271', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.73385620', 'grpo_loop_time': '419.53417969', 'ref_model_time': '112.35651398', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84320349', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.21923828', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.489e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.00752666', 'kl_moving': '0.00752666', 'kl_stepwise': '-2.001e-08', 'gathered_avg_reward_mean': '9.22253418', 'gathered_avg_reward_std': '1.25703895', 'gathered_video_score2_mq_reward_mean': '3.18676758', 'gathered_video_score2_mq_reward_std': '0.43133941', 'gathered_video_score2_overall_reward_mean': '9.21923828', 'gathered_video_score2_overall_reward_std': '1.25775778', 'gathered_video_score2_ta_reward_mean': '2.95507812', 'gathered_video_score2_ta_reward_std': '0.67799181', 'gathered_video_score2_vq_reward_mean': '3.08068848', 'gathered_video_score2_vq_reward_std': '0.58093417', 'gathered_ori_avg_reward_mean': '9.22253418', 'gathered_ori_avg_reward_std': '1.25703895'} +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 11:53:36.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_0/index_8076_seed_31063_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 11:55:02.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:55:02.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:07.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_0/index_8076_seed_31064_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 11:56:30.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:56:30.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 11:56:31.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.056 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 178 train_steps 3120 {'step': '178/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-10', 'grad_norm': '0.00538353', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.87457275', 'grpo_loop_time': '418.87326050', 'ref_model_time': '112.68727112', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87260628', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.80126953', 'token_credit_video_advantage_mean': '3.492e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.409e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.260e-09', 'kl_loss': '0.00813719', 'kl_moving': '0.00813719', 'kl_stepwise': '-1.889e-08', 'gathered_avg_reward_mean': '8.80224609', 'gathered_avg_reward_std': '0.88196182', 'gathered_video_score2_mq_reward_mean': '3.05895996', 'gathered_video_score2_mq_reward_std': '0.36153904', 'gathered_video_score2_overall_reward_mean': '8.80126953', 'gathered_video_score2_overall_reward_std': '0.88195419', 'gathered_video_score2_ta_reward_mean': '2.89892578', 'gathered_video_score2_ta_reward_std': '0.59844542', 'gathered_video_score2_vq_reward_mean': '2.84436035', 'gathered_video_score2_vq_reward_std': '0.53970838', 'gathered_ori_avg_reward_mean': '8.80224609', 'gathered_ori_avg_reward_std': '0.88196176'} +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 12:07:21.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:27.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_0/index_2217_seed_24417_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 12:08:41.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:41.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:09:51.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_0/index_2217_seed_24418_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 12:10:11.286 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:10:11.454 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 12:10:12.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.684 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 179 train_steps 3160 {'step': '179/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00505435', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.44982910', 'grpo_loop_time': '419.96017456', 'ref_model_time': '113.61384583', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87367129', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.86572266', 'token_credit_video_advantage_mean': '6.054e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.441e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.794e-09', 'kl_loss': '0.00837230', 'kl_moving': '0.00837230', 'kl_stepwise': '-1.935e-08', 'gathered_avg_reward_mean': '8.86340332', 'gathered_avg_reward_std': '0.93295610', 'gathered_video_score2_mq_reward_mean': '3.10937500', 'gathered_video_score2_mq_reward_std': '0.26336065', 'gathered_video_score2_overall_reward_mean': '8.86572266', 'gathered_video_score2_overall_reward_std': '0.93785280', 'gathered_video_score2_ta_reward_mean': '2.92053223', 'gathered_video_score2_ta_reward_std': '0.50337523', 'gathered_video_score2_vq_reward_mean': '2.83349609', 'gathered_video_score2_vq_reward_std': '0.55682713', 'gathered_ori_avg_reward_mean': '8.86340332', 'gathered_ori_avg_reward_std': '0.93295610'} +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 12:21:06.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:13.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_0/index_2237_seed_22609_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 12:22:27.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:27.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:39.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_0/index_2237_seed_22610_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:23:59.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 12:24:00.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:34:53.001 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 180 train_steps 3200 {'step': '180/10000', 'learning_rate': '0.00001000', 'loss': '-1.024e-08', 'grad_norm': '0.00570710', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.79101562', 'grpo_loop_time': '420.40637207', 'ref_model_time': '112.90185547', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86158615', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.50976562', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.397e-08', 'kl_loss': '0.00820204', 'kl_moving': '0.00820204', 'kl_stepwise': '-1.949e-08', 'gathered_avg_reward_mean': '8.51354980', 'gathered_avg_reward_std': '1.25040913', 'gathered_video_score2_mq_reward_mean': '3.02563477', 'gathered_video_score2_mq_reward_std': '0.44623342', 'gathered_video_score2_overall_reward_mean': '8.50976562', 'gathered_video_score2_overall_reward_std': '1.24974215', 'gathered_video_score2_ta_reward_mean': '2.67358398', 'gathered_video_score2_ta_reward_std': '0.58854640', 'gathered_video_score2_vq_reward_mean': '2.81433105', 'gathered_video_score2_vq_reward_std': '0.57262170', 'gathered_ori_avg_reward_mean': '8.51354980', 'gathered_ori_avg_reward_std': '1.25040901'} +2026-06-26 12:35:29.017 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 12:35:30.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:38.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_0/index_357_seed_1772_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 12:36:55.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:55.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:04.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_0/index_357_seed_1773_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:38:24.487 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 12:38:25.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.047 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 181 train_steps 3240 {'step': '181/10000', 'learning_rate': '0.00001000', 'loss': '-1.770e-08', 'grad_norm': '0.00874148', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.22906494', 'grpo_loop_time': '418.15875244', 'ref_model_time': '112.61692047', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.90125823', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.85839844', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.433e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.770e-08', 'kl_loss': '0.00755630', 'kl_moving': '0.00755630', 'kl_stepwise': '-1.947e-08', 'gathered_avg_reward_mean': '8.85742188', 'gathered_avg_reward_std': '1.11854911', 'gathered_video_score2_mq_reward_mean': '3.10485840', 'gathered_video_score2_mq_reward_std': '0.39405188', 'gathered_video_score2_overall_reward_mean': '8.85839844', 'gathered_video_score2_overall_reward_std': '1.11975193', 'gathered_video_score2_ta_reward_mean': '2.97326660', 'gathered_video_score2_ta_reward_std': '0.57566321', 'gathered_video_score2_vq_reward_mean': '2.77929688', 'gathered_video_score2_vq_reward_std': '0.52106357', 'gathered_ori_avg_reward_mean': '8.85742188', 'gathered_ori_avg_reward_std': '1.11854911'} +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 12:49:16.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:22.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_0/index_6482_seed_17570_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 12:50:38.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:38.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:51:48.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_0/index_6482_seed_17571_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:52:09.209 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 12:52:10.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.728 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 182 train_steps 3280 {'step': '182/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00567619', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.63427734', 'grpo_loop_time': '421.12667847', 'ref_model_time': '112.94380951', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86438942', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03466797', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.457e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.794e-09', 'kl_loss': '0.00730811', 'kl_moving': '0.00730811', 'kl_stepwise': '-1.923e-08', 'gathered_avg_reward_mean': '9.03063965', 'gathered_avg_reward_std': '0.96309990', 'gathered_video_score2_mq_reward_mean': '3.15087891', 'gathered_video_score2_mq_reward_std': '0.35274562', 'gathered_video_score2_overall_reward_mean': '9.03466797', 'gathered_video_score2_overall_reward_std': '0.96334994', 'gathered_video_score2_ta_reward_mean': '3.03601074', 'gathered_video_score2_ta_reward_std': '0.60358614', 'gathered_video_score2_vq_reward_mean': '2.84375000', 'gathered_video_score2_vq_reward_std': '0.50972635', 'gathered_ori_avg_reward_mean': '9.03063965', 'gathered_ori_avg_reward_std': '0.96309990'} +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 13:03:03.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:09.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_0/index_4776_seed_6618_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 13:04:22.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:22.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:33.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_0/index_4776_seed_6619_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:05:52.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 13:05:53.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.884 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 183 train_steps 3320 {'step': '183/10000', 'learning_rate': '0.00001000', 'loss': '2.095e-09', 'grad_norm': '0.00636480', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.90972900', 'grpo_loop_time': '419.12847900', 'ref_model_time': '112.89694977', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86542624', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.69580078', 'token_credit_video_advantage_mean': '6.286e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.422e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.164e-09', 'kl_loss': '0.00741098', 'kl_moving': '0.00741098', 'kl_stepwise': '-1.993e-08', 'gathered_avg_reward_mean': '8.69409180', 'gathered_avg_reward_std': '1.37786782', 'gathered_video_score2_mq_reward_mean': '3.05029297', 'gathered_video_score2_mq_reward_std': '0.49551070', 'gathered_video_score2_overall_reward_mean': '8.69580078', 'gathered_video_score2_overall_reward_std': '1.37882590', 'gathered_video_score2_ta_reward_mean': '2.82360840', 'gathered_video_score2_ta_reward_std': '0.68333817', 'gathered_video_score2_vq_reward_mean': '2.82019043', 'gathered_video_score2_vq_reward_std': '0.54064798', 'gathered_ori_avg_reward_mean': '8.69409180', 'gathered_ori_avg_reward_std': '1.37786794'} +2026-06-26 13:16:43.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 13:16:44.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_0/index_306_seed_30904_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 13:18:09.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:18:09.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:17.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_0/index_306_seed_30905_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:19:40.749 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 13:19:41.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:30.949 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 184 train_steps 3360 {'step': '184/10000', 'learning_rate': '0.00001000', 'loss': '2.072e-08', 'grad_norm': '0.00489249', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.36184692', 'grpo_loop_time': '417.61105347', 'ref_model_time': '112.71751404', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85873842', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14111328', 'token_credit_video_advantage_mean': '-5.355e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.497e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.049e-08', 'kl_loss': '0.00715506', 'kl_moving': '0.00715506', 'kl_stepwise': '-1.930e-08', 'gathered_avg_reward_mean': '9.13623047', 'gathered_avg_reward_std': '1.08450055', 'gathered_video_score2_mq_reward_mean': '3.14440918', 'gathered_video_score2_mq_reward_std': '0.40662307', 'gathered_video_score2_overall_reward_mean': '9.14111328', 'gathered_video_score2_overall_reward_std': '1.08541739', 'gathered_video_score2_ta_reward_mean': '2.99035645', 'gathered_video_score2_ta_reward_std': '0.63625687', 'gathered_video_score2_vq_reward_mean': '3.00146484', 'gathered_video_score2_vq_reward_std': '0.49772677', 'gathered_ori_avg_reward_mean': '9.13623047', 'gathered_ori_avg_reward_std': '1.08450055'} +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 13:30:31.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:37.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_0/index_7318_seed_27899_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 13:31:54.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:54.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:05.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_0/index_7318_seed_27900_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:33:24.718 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 13:33:25.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:14.988 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 185 train_steps 3400 {'step': '185/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00444932', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.63088989', 'grpo_loop_time': '418.20358276', 'ref_model_time': '112.33097076', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85434127', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.72656250', 'token_credit_video_advantage_mean': '4.424e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.455e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.00755112', 'kl_moving': '0.00755113', 'kl_stepwise': '-1.983e-08', 'gathered_avg_reward_mean': '8.72497559', 'gathered_avg_reward_std': '1.03060353', 'gathered_video_score2_mq_reward_mean': '3.01867676', 'gathered_video_score2_mq_reward_std': '0.32685423', 'gathered_video_score2_overall_reward_mean': '8.72656250', 'gathered_video_score2_overall_reward_std': '1.02849424', 'gathered_video_score2_ta_reward_mean': '2.98852539', 'gathered_video_score2_ta_reward_std': '0.63328248', 'gathered_video_score2_vq_reward_mean': '2.71777344', 'gathered_video_score2_vq_reward_std': '0.49419564', 'gathered_ori_avg_reward_mean': '8.72497559', 'gathered_ori_avg_reward_std': '1.03060353'} +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 13:44:15.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:21.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_0/index_7637_seed_15724_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 13:45:40.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:40.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:46:49.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_0/index_7637_seed_15725_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:47:12.877 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 13:47:13.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.225 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 186 train_steps 3440 {'step': '186/10000', 'learning_rate': '0.00001000', 'loss': '3.260e-09', 'grad_norm': '0.00475813', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '293.01675415', 'grpo_loop_time': '419.91921997', 'ref_model_time': '113.78520203', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84250265', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.36572266', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.473e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.260e-09', 'kl_loss': '0.00815576', 'kl_moving': '0.00815576', 'kl_stepwise': '-2.015e-08', 'gathered_avg_reward_mean': '9.37231445', 'gathered_avg_reward_std': '0.99255103', 'gathered_video_score2_mq_reward_mean': '3.22167969', 'gathered_video_score2_mq_reward_std': '0.27414632', 'gathered_video_score2_overall_reward_mean': '9.36572266', 'gathered_video_score2_overall_reward_std': '0.99008882', 'gathered_video_score2_ta_reward_mean': '3.24963379', 'gathered_video_score2_ta_reward_std': '0.59566700', 'gathered_video_score2_vq_reward_mean': '2.90100098', 'gathered_video_score2_vq_reward_std': '0.40352255', 'gathered_ori_avg_reward_mean': '9.37231445', 'gathered_ori_avg_reward_std': '0.99255097'} +2026-06-26 13:58:07.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 13:58:07.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:13.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_0/index_7339_seed_10900_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 13:59:27.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:27.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:36.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_0/index_7339_seed_10901_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:00:56.135 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 14:00:56.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.267 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 187 train_steps 3480 {'step': '187/10000', 'learning_rate': '0.00001000', 'loss': '1.397e-09', 'grad_norm': '0.00666219', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.89770508', 'grpo_loop_time': '418.52737427', 'ref_model_time': '114.44575500', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88140798', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.88720703', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-09', 'kl_loss': '0.00917839', 'kl_moving': '0.00917839', 'kl_stepwise': '-1.970e-08', 'gathered_avg_reward_mean': '8.88842773', 'gathered_avg_reward_std': '1.11584771', 'gathered_video_score2_mq_reward_mean': '3.15197754', 'gathered_video_score2_mq_reward_std': '0.33553678', 'gathered_video_score2_overall_reward_mean': '8.88720703', 'gathered_video_score2_overall_reward_std': '1.11671782', 'gathered_video_score2_ta_reward_mean': '2.92980957', 'gathered_video_score2_ta_reward_std': '0.70754027', 'gathered_video_score2_vq_reward_mean': '2.80664062', 'gathered_video_score2_vq_reward_std': '0.55040538', 'gathered_ori_avg_reward_mean': '8.88842773', 'gathered_ori_avg_reward_std': '1.11584771'} +2026-06-26 14:11:49.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 14:11:49.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:55.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_0/index_1372_seed_1463_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 14:13:12.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:13:12.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:18.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_0/index_1372_seed_1464_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:14:38.335 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 14:14:39.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.784 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 188 train_steps 3520 {'step': '188/10000', 'learning_rate': '0.00001000', 'loss': '2.328e-09', 'grad_norm': '0.00800138', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.50109863', 'grpo_loop_time': '417.54681396', 'ref_model_time': '112.89270782', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86655176', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.50537109', 'token_credit_video_advantage_mean': '-3.492e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.479e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.328e-09', 'kl_loss': '0.00913852', 'kl_moving': '0.00913853', 'kl_stepwise': '-2.009e-08', 'gathered_avg_reward_mean': '8.50573730', 'gathered_avg_reward_std': '0.99419355', 'gathered_video_score2_mq_reward_mean': '3.03063965', 'gathered_video_score2_mq_reward_std': '0.28135952', 'gathered_video_score2_overall_reward_mean': '8.50537109', 'gathered_video_score2_overall_reward_std': '0.99568093', 'gathered_video_score2_ta_reward_mean': '2.87524414', 'gathered_video_score2_ta_reward_std': '0.47439426', 'gathered_video_score2_vq_reward_mean': '2.59985352', 'gathered_video_score2_vq_reward_std': '0.49996084', 'gathered_ori_avg_reward_mean': '8.50573730', 'gathered_ori_avg_reward_std': '0.99419361'} +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 14:25:27.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:34.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_0/index_8870_seed_20304_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 14:26:47.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:47.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:02.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_0/index_8870_seed_20305_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 14:28:24.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:28:24.424 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 14:28:25.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.309 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 189 train_steps 3560 {'step': '189/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00487655', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.72705078', 'grpo_loop_time': '418.11059570', 'ref_model_time': '112.58667755', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86026675', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.92089844', 'token_credit_video_advantage_mean': '8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.472e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.304e-08', 'kl_loss': '0.00850891', 'kl_moving': '0.00850891', 'kl_stepwise': '-1.997e-08', 'gathered_avg_reward_mean': '8.91650391', 'gathered_avg_reward_std': '1.15074527', 'gathered_video_score2_mq_reward_mean': '3.11755371', 'gathered_video_score2_mq_reward_std': '0.42443845', 'gathered_video_score2_overall_reward_mean': '8.92089844', 'gathered_video_score2_overall_reward_std': '1.15119743', 'gathered_video_score2_ta_reward_mean': '2.89196777', 'gathered_video_score2_ta_reward_std': '0.57565778', 'gathered_video_score2_vq_reward_mean': '2.90698242', 'gathered_video_score2_vq_reward_std': '0.47363326', 'gathered_ori_avg_reward_mean': '8.91650391', 'gathered_ori_avg_reward_std': '1.15074527'} +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 14:39:15.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_0/index_3239_seed_18003_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 14:40:41.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:41.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:41:50.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_0/index_3239_seed_18004_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 14:42:12.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:42:12.268 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 14:42:13.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:07.085 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 190 train_steps 3600 {'step': '190/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00477126', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.49218750', 'grpo_loop_time': '420.08642578', 'ref_model_time': '114.00801849', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.82399815', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.11572266', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00753981', 'kl_moving': '0.00753981', 'kl_stepwise': '-1.941e-08', 'gathered_avg_reward_mean': '9.11767578', 'gathered_avg_reward_std': '1.30517054', 'gathered_video_score2_mq_reward_mean': '3.17333984', 'gathered_video_score2_mq_reward_std': '0.42620856', 'gathered_video_score2_overall_reward_mean': '9.11572266', 'gathered_video_score2_overall_reward_std': '1.30522168', 'gathered_video_score2_ta_reward_mean': '2.90222168', 'gathered_video_score2_ta_reward_std': '0.71556377', 'gathered_video_score2_vq_reward_mean': '3.04211426', 'gathered_video_score2_vq_reward_std': '0.51763457', 'gathered_ori_avg_reward_mean': '9.11767578', 'gathered_ori_avg_reward_std': '1.30517054'} +2026-06-26 14:53:42.630 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 14:53:45.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:53.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_0/index_9416_seed_10361_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 14:55:06.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:55:06.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:17.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_0/index_9416_seed_10362_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:56:35.746 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 14:56:36.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.253 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 191 train_steps 3640 {'step': '191/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00394124', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.54498291', 'grpo_loop_time': '419.27413940', 'ref_model_time': '112.11903381', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84299839', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14013672', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00685068', 'kl_moving': '0.00685068', 'kl_stepwise': '-1.846e-08', 'gathered_avg_reward_mean': '9.13891602', 'gathered_avg_reward_std': '1.02687991', 'gathered_video_score2_mq_reward_mean': '3.15356445', 'gathered_video_score2_mq_reward_std': '0.33866933', 'gathered_video_score2_overall_reward_mean': '9.14013672', 'gathered_video_score2_overall_reward_std': '1.02625144', 'gathered_video_score2_ta_reward_mean': '3.08642578', 'gathered_video_score2_ta_reward_std': '0.54894954', 'gathered_video_score2_vq_reward_mean': '2.89892578', 'gathered_video_score2_vq_reward_std': '0.45718122', 'gathered_ori_avg_reward_mean': '9.13891602', 'gathered_ori_avg_reward_std': '1.02687991'} +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 15:07:27.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:33.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_0/index_827_seed_17315_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 15:08:48.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:48.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:09:56.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_0/index_827_seed_17316_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:10:19.032 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 15:10:19.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.758 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 192 train_steps 3680 {'step': '192/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00530738', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.21417236', 'grpo_loop_time': '419.46514893', 'ref_model_time': '111.98398590', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87066025', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.69189453', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.436e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00743718', 'kl_moving': '0.00743719', 'kl_stepwise': '-1.896e-08', 'gathered_avg_reward_mean': '8.69372559', 'gathered_avg_reward_std': '1.32813156', 'gathered_video_score2_mq_reward_mean': '3.05273438', 'gathered_video_score2_mq_reward_std': '0.44894481', 'gathered_video_score2_overall_reward_mean': '8.69189453', 'gathered_video_score2_overall_reward_std': '1.32683611', 'gathered_video_score2_ta_reward_mean': '2.82897949', 'gathered_video_score2_ta_reward_std': '0.67133272', 'gathered_video_score2_vq_reward_mean': '2.81201172', 'gathered_video_score2_vq_reward_std': '0.52239406', 'gathered_ori_avg_reward_mean': '8.69372559', 'gathered_ori_avg_reward_std': '1.32813156'} +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 15:21:10.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:17.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_0/index_6760_seed_17098_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 15:22:35.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:35.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:23:41.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_0/index_6760_seed_17099_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 15:24:01.111 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:24:01.252 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 15:24:02.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.901 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 193 train_steps 3720 {'step': '193/10000', 'learning_rate': '0.00001000', 'loss': '-9.686e-09', 'grad_norm': '0.00491134', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.24899292', 'grpo_loop_time': '421.91369629', 'ref_model_time': '112.86131287', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88602465', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.06933594', 'token_credit_video_advantage_mean': '-9.779e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.441e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.024e-08', 'kl_loss': '0.00723153', 'kl_moving': '0.00723153', 'kl_stepwise': '-1.931e-08', 'gathered_avg_reward_mean': '9.07556152', 'gathered_avg_reward_std': '1.03377163', 'gathered_video_score2_mq_reward_mean': '3.16162109', 'gathered_video_score2_mq_reward_std': '0.32640243', 'gathered_video_score2_overall_reward_mean': '9.06933594', 'gathered_video_score2_overall_reward_std': '1.03443325', 'gathered_video_score2_ta_reward_mean': '3.01184082', 'gathered_video_score2_ta_reward_std': '0.61041641', 'gathered_video_score2_vq_reward_mean': '2.90209961', 'gathered_video_score2_vq_reward_std': '0.47926235', 'gathered_ori_avg_reward_mean': '9.07556152', 'gathered_ori_avg_reward_std': '1.03377163'} +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 15:34:56.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:02.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_0/index_4904_seed_17667_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 15:36:17.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:17.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:30.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_0/index_4904_seed_17668_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:37:49.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 15:37:50.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.212 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 194 train_steps 3760 {'step': '194/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00459128', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.22640991', 'grpo_loop_time': '421.27960205', 'ref_model_time': '112.94021606', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86764926', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20263672', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.478e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00732346', 'kl_moving': '0.00732346', 'kl_stepwise': '-1.998e-08', 'gathered_avg_reward_mean': '9.20239258', 'gathered_avg_reward_std': '0.71390754', 'gathered_video_score2_mq_reward_mean': '3.14697266', 'gathered_video_score2_mq_reward_std': '0.26873353', 'gathered_video_score2_overall_reward_mean': '9.20263672', 'gathered_video_score2_overall_reward_std': '0.71370065', 'gathered_video_score2_ta_reward_mean': '3.14147949', 'gathered_video_score2_ta_reward_std': '0.56146502', 'gathered_video_score2_vq_reward_mean': '2.91394043', 'gathered_video_score2_vq_reward_std': '0.41035962', 'gathered_ori_avg_reward_mean': '9.20239258', 'gathered_ori_avg_reward_std': '0.71390754'} +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 15:48:45.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:51.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_0/index_2342_seed_31533_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 15:50:09.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:50:09.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:15.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_0/index_2342_seed_31534_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:51:34.814 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 15:51:35.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.541 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 195 train_steps 3800 {'step': '195/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00482950', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.00683594', 'grpo_loop_time': '419.51007080', 'ref_model_time': '112.78923035', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87205738', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.69335938', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.442e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.00750656', 'kl_moving': '0.00750656', 'kl_stepwise': '-1.914e-08', 'gathered_avg_reward_mean': '8.69335938', 'gathered_avg_reward_std': '1.34402883', 'gathered_video_score2_mq_reward_mean': '2.98107910', 'gathered_video_score2_mq_reward_std': '0.38641977', 'gathered_video_score2_overall_reward_mean': '8.69335938', 'gathered_video_score2_overall_reward_std': '1.34394157', 'gathered_video_score2_ta_reward_mean': '2.96533203', 'gathered_video_score2_ta_reward_std': '0.73256361', 'gathered_video_score2_vq_reward_mean': '2.74694824', 'gathered_video_score2_vq_reward_std': '0.50797141', 'gathered_ori_avg_reward_mean': '8.69335938', 'gathered_ori_avg_reward_std': '1.34402883'} +2026-06-26 16:02:26.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 16:02:26.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_0/index_6549_seed_31222_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 16:03:46.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:46.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:04:57.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_0/index_6549_seed_31223_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 16:05:16.921 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:05:17.090 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 16:05:18.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.653 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 196 train_steps 3840 {'step': '196/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00755416', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.46398926', 'grpo_loop_time': '417.71685791', 'ref_model_time': '112.22235870', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85176528', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.80224609', 'token_credit_video_advantage_mean': '-1.024e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.426e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00785360', 'kl_moving': '0.00785360', 'kl_stepwise': '-1.979e-08', 'gathered_avg_reward_mean': '8.80212402', 'gathered_avg_reward_std': '0.85033989', 'gathered_video_score2_mq_reward_mean': '3.12182617', 'gathered_video_score2_mq_reward_std': '0.22524841', 'gathered_video_score2_overall_reward_mean': '8.80224609', 'gathered_video_score2_overall_reward_std': '0.84937459', 'gathered_video_score2_ta_reward_mean': '2.81506348', 'gathered_video_score2_ta_reward_std': '0.54459459', 'gathered_video_score2_vq_reward_mean': '2.86523438', 'gathered_video_score2_vq_reward_std': '0.51007420', 'gathered_ori_avg_reward_mean': '8.80212402', 'gathered_ori_avg_reward_std': '0.85033989'} +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 16:16:06.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:13.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_0/index_8016_seed_1341_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 16:17:29.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:29.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:18:42.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_0/index_8016_seed_1342_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:19:03.185 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 16:19:04.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.543 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 197 train_steps 3880 {'step': '197/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00526815', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.77117920', 'grpo_loop_time': '418.43902588', 'ref_model_time': '112.53258514', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89004409', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.37890625', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.415e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00813122', 'kl_moving': '0.00813122', 'kl_stepwise': '-1.924e-08', 'gathered_avg_reward_mean': '9.37255859', 'gathered_avg_reward_std': '1.04225242', 'gathered_video_score2_mq_reward_mean': '3.20971680', 'gathered_video_score2_mq_reward_std': '0.41279662', 'gathered_video_score2_overall_reward_mean': '9.37890625', 'gathered_video_score2_overall_reward_std': '1.03933418', 'gathered_video_score2_ta_reward_mean': '3.26281738', 'gathered_video_score2_ta_reward_std': '0.55281436', 'gathered_video_score2_vq_reward_mean': '2.90002441', 'gathered_video_score2_vq_reward_std': '0.50376844', 'gathered_ori_avg_reward_mean': '9.37255859', 'gathered_ori_avg_reward_std': '1.04225242'} +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 16:29:54.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_0/index_5950_seed_24152_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 16:31:15.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:15.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:24.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_0/index_5950_seed_24153_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:32:45.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 16:32:45.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.037 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 198 train_steps 3920 {'step': '198/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00564939', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.48239136', 'grpo_loop_time': '418.96972656', 'ref_model_time': '112.29309845', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86935544', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.08349609', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.369e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00757058', 'kl_moving': '0.00757059', 'kl_stepwise': '-1.895e-08', 'gathered_avg_reward_mean': '9.08312988', 'gathered_avg_reward_std': '1.34623206', 'gathered_video_score2_mq_reward_mean': '3.09265137', 'gathered_video_score2_mq_reward_std': '0.43549842', 'gathered_video_score2_overall_reward_mean': '9.08349609', 'gathered_video_score2_overall_reward_std': '1.34742475', 'gathered_video_score2_ta_reward_mean': '2.93054199', 'gathered_video_score2_ta_reward_std': '0.68599069', 'gathered_video_score2_vq_reward_mean': '3.05993652', 'gathered_video_score2_vq_reward_std': '0.48192355', 'gathered_ori_avg_reward_mean': '9.08312988', 'gathered_ori_avg_reward_std': '1.34623194'} +2026-06-26 16:43:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 16:43:38.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:44.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_0/index_6234_seed_10501_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 16:44:59.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:59.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:08.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_0/index_6234_seed_10502_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:46:30.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 16:46:31.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.622 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 199 train_steps 3960 {'step': '199/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00546153', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.36413574', 'grpo_loop_time': '417.37686157', 'ref_model_time': '113.02416229', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87406838', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.12402344', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00812082', 'kl_moving': '0.00812082', 'kl_stepwise': '-1.997e-08', 'gathered_avg_reward_mean': '9.12561035', 'gathered_avg_reward_std': '0.99144638', 'gathered_video_score2_mq_reward_mean': '3.11389160', 'gathered_video_score2_mq_reward_std': '0.36873686', 'gathered_video_score2_overall_reward_mean': '9.12402344', 'gathered_video_score2_overall_reward_std': '0.99312466', 'gathered_video_score2_ta_reward_mean': '3.01098633', 'gathered_video_score2_ta_reward_std': '0.49057803', 'gathered_video_score2_vq_reward_mean': '3.00073242', 'gathered_video_score2_vq_reward_std': '0.54698783', 'gathered_ori_avg_reward_mean': '9.12561035', 'gathered_ori_avg_reward_std': '0.99144644'} +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 16:57:20.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_0/index_9805_seed_298_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 16:58:41.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:41.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:59:54.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_0/index_9805_seed_299_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:00:15.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 17:00:15.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:11:06.993 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 200 train_steps 4000 {'step': '200/10000', 'learning_rate': '0.00001000', 'loss': '-1.467e-08', 'grad_norm': '0.00515233', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.33074951', 'grpo_loop_time': '418.94677734', 'ref_model_time': '113.36005402', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89133173', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07861328', 'token_credit_video_advantage_mean': '-1.164e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.412e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.467e-08', 'kl_loss': '0.00865205', 'kl_moving': '0.00865205', 'kl_stepwise': '-1.943e-08', 'gathered_avg_reward_mean': '9.07897949', 'gathered_avg_reward_std': '1.18300152', 'gathered_video_score2_mq_reward_mean': '3.16870117', 'gathered_video_score2_mq_reward_std': '0.39625993', 'gathered_video_score2_overall_reward_mean': '9.07861328', 'gathered_video_score2_overall_reward_std': '1.18197465', 'gathered_video_score2_ta_reward_mean': '2.97863770', 'gathered_video_score2_ta_reward_std': '0.57085609', 'gathered_video_score2_vq_reward_mean': '2.93164062', 'gathered_video_score2_vq_reward_std': '0.50413901', 'gathered_ori_avg_reward_mean': '9.07897949', 'gathered_ori_avg_reward_std': '1.18300152'} +2026-06-26 17:12:12.739 | INFO | __main__:save_checkpoint:1967 - Checkpoint saved at step 200 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-200 +2026-06-26 17:12:12.870 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 200: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-200 +2026-06-26 17:12:15.622 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-175 +2026-06-26 17:12:51.094 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 17:12:53.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:14:00.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_0/index_2391_seed_20199_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 17:14:15.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:14:15.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:26.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_0/index_2391_seed_20200_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:15:47.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 17:15:48.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.454 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 201 train_steps 4040 {'step': '201/10000', 'learning_rate': '0.00001000', 'loss': '-3.260e-09', 'grad_norm': '0.00475470', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.21228027', 'grpo_loop_time': '419.60446167', 'ref_model_time': '112.20094299', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86125565', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.99267578', 'token_credit_video_advantage_mean': '-1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.446e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.260e-09', 'kl_loss': '0.00752522', 'kl_moving': '0.00752522', 'kl_stepwise': '-2.013e-08', 'gathered_avg_reward_mean': '8.99011230', 'gathered_avg_reward_std': '1.16564775', 'gathered_video_score2_mq_reward_mean': '3.14733887', 'gathered_video_score2_mq_reward_std': '0.40832195', 'gathered_video_score2_overall_reward_mean': '8.99267578', 'gathered_video_score2_overall_reward_std': '1.16692626', 'gathered_video_score2_ta_reward_mean': '2.90759277', 'gathered_video_score2_ta_reward_std': '0.54995221', 'gathered_video_score2_vq_reward_mean': '2.93518066', 'gathered_video_score2_vq_reward_std': '0.55162156', 'gathered_ori_avg_reward_mean': '8.99011230', 'gathered_ori_avg_reward_std': '1.16564775'} +2026-06-26 17:26:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 17:26:39.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:45.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_0/index_6886_seed_28497_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 17:28:00.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:28:00.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:09.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_0/index_6886_seed_28498_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:29:29.559 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 17:29:30.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.051 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 202 train_steps 4080 {'step': '202/10000', 'learning_rate': '0.00001000', 'loss': '-2.049e-08', 'grad_norm': '0.00617777', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.38031006', 'grpo_loop_time': '418.28799438', 'ref_model_time': '112.85041046', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88223803', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01416016', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.049e-08', 'kl_loss': '0.00767993', 'kl_moving': '0.00767993', 'kl_stepwise': '-1.914e-08', 'gathered_avg_reward_mean': '9.00854492', 'gathered_avg_reward_std': '1.22776449', 'gathered_video_score2_mq_reward_mean': '3.12414551', 'gathered_video_score2_mq_reward_std': '0.42921254', 'gathered_video_score2_overall_reward_mean': '9.01416016', 'gathered_video_score2_overall_reward_std': '1.23139143', 'gathered_video_score2_ta_reward_mean': '2.97155762', 'gathered_video_score2_ta_reward_std': '0.61535740', 'gathered_video_score2_vq_reward_mean': '2.91284180', 'gathered_video_score2_vq_reward_std': '0.53522080', 'gathered_ori_avg_reward_mean': '9.00854492', 'gathered_ori_avg_reward_std': '1.22776449'} +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 17:40:20.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:26.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_0/index_6375_seed_22232_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 17:41:40.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:40.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:42:54.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_0/index_6375_seed_22233_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 17:43:13.842 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:43:13.966 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 17:43:14.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.534 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 203 train_steps 4120 {'step': '203/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00680819', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.98452759', 'grpo_loop_time': '418.91503906', 'ref_model_time': '113.51608276', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86600912', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.79882812', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.474e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00811994', 'kl_moving': '0.00811994', 'kl_stepwise': '-1.989e-08', 'gathered_avg_reward_mean': '8.79846191', 'gathered_avg_reward_std': '1.01957250', 'gathered_video_score2_mq_reward_mean': '3.03955078', 'gathered_video_score2_mq_reward_std': '0.36002159', 'gathered_video_score2_overall_reward_mean': '8.79882812', 'gathered_video_score2_overall_reward_std': '1.02070355', 'gathered_video_score2_ta_reward_mean': '2.96252441', 'gathered_video_score2_ta_reward_std': '0.43657517', 'gathered_video_score2_vq_reward_mean': '2.79638672', 'gathered_video_score2_vq_reward_std': '0.46988973', 'gathered_ori_avg_reward_mean': '8.79846191', 'gathered_ori_avg_reward_std': '1.01957262'} +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 17:54:06.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:12.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_0/index_6280_seed_31269_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 17:55:29.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:29.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:38.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_0/index_6280_seed_31270_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:56:58.134 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 17:56:59.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.312 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 204 train_steps 4160 {'step': '204/10000', 'learning_rate': '0.00001000', 'loss': '-1.397e-08', 'grad_norm': '0.00469059', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.03280640', 'grpo_loop_time': '418.02014160', 'ref_model_time': '113.93004608', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84133720', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.97167969', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.429e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.397e-08', 'kl_loss': '0.00764150', 'kl_moving': '0.00764150', 'kl_stepwise': '-1.957e-08', 'gathered_avg_reward_mean': '8.97082520', 'gathered_avg_reward_std': '0.94627166', 'gathered_video_score2_mq_reward_mean': '3.16210938', 'gathered_video_score2_mq_reward_std': '0.28757402', 'gathered_video_score2_overall_reward_mean': '8.97167969', 'gathered_video_score2_overall_reward_std': '0.94574368', 'gathered_video_score2_ta_reward_mean': '2.83544922', 'gathered_video_score2_ta_reward_std': '0.55621362', 'gathered_video_score2_vq_reward_mean': '2.97326660', 'gathered_video_score2_vq_reward_std': '0.52865469', 'gathered_ori_avg_reward_mean': '8.97082520', 'gathered_ori_avg_reward_std': '0.94627166'} +2026-06-26 18:07:49.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 18:07:49.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:55.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_0/index_9442_seed_23321_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 18:09:10.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:09:10.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:21.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_0/index_9442_seed_23322_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 18:10:44.980 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:10:45.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 18:10:46.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.495 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 205 train_steps 4200 {'step': '205/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00635198', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.59356689', 'grpo_loop_time': '419.61306763', 'ref_model_time': '112.14122772', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85107273', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.54980469', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.421e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00786240', 'kl_moving': '0.00786240', 'kl_stepwise': '-1.930e-08', 'gathered_avg_reward_mean': '9.54785156', 'gathered_avg_reward_std': '1.04307342', 'gathered_video_score2_mq_reward_mean': '3.15039062', 'gathered_video_score2_mq_reward_std': '0.32043752', 'gathered_video_score2_overall_reward_mean': '9.54980469', 'gathered_video_score2_overall_reward_std': '1.04531014', 'gathered_video_score2_ta_reward_mean': '3.43969727', 'gathered_video_score2_ta_reward_std': '0.64050955', 'gathered_video_score2_vq_reward_mean': '2.95776367', 'gathered_video_score2_vq_reward_std': '0.46076101', 'gathered_ori_avg_reward_mean': '9.54785156', 'gathered_ori_avg_reward_std': '1.04307342'} +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 18:21:38.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:44.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_0/index_7083_seed_17328_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 18:23:00.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:23:00.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:12.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_0/index_7083_seed_17329_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:24:36.360 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 18:24:37.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.617 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 206 train_steps 4240 {'step': '206/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00639535', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.16714478', 'grpo_loop_time': '418.92773438', 'ref_model_time': '112.80796814', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86907172', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.72070312', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.456e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.304e-08', 'kl_loss': '0.00785075', 'kl_moving': '0.00785076', 'kl_stepwise': '-2.006e-08', 'gathered_avg_reward_mean': '8.71997070', 'gathered_avg_reward_std': '1.62322330', 'gathered_video_score2_mq_reward_mean': '3.06652832', 'gathered_video_score2_mq_reward_std': '0.51832384', 'gathered_video_score2_overall_reward_mean': '8.72070312', 'gathered_video_score2_overall_reward_std': '1.62227345', 'gathered_video_score2_ta_reward_mean': '2.90490723', 'gathered_video_score2_ta_reward_std': '0.71804714', 'gathered_video_score2_vq_reward_mean': '2.74853516', 'gathered_video_score2_vq_reward_std': '0.58597797', 'gathered_ori_avg_reward_mean': '8.71997070', 'gathered_ori_avg_reward_std': '1.62322330'} +2026-06-26 18:35:27.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 18:35:27.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:34.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_0/index_6976_seed_24611_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 18:36:51.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:51.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:04.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_0/index_6976_seed_24612_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:38:23.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 18:38:24.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:14.997 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 207 train_steps 4280 {'step': '207/10000', 'learning_rate': '0.00001000', 'loss': '8.382e-09', 'grad_norm': '0.00679746', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.68191528', 'grpo_loop_time': '418.75900269', 'ref_model_time': '112.68565369', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89606905', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.58007812', 'token_credit_video_advantage_mean': '-5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.429e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00783930', 'kl_moving': '0.00783931', 'kl_stepwise': '-1.894e-08', 'gathered_avg_reward_mean': '9.57812500', 'gathered_avg_reward_std': '1.34968567', 'gathered_video_score2_mq_reward_mean': '3.28930664', 'gathered_video_score2_mq_reward_std': '0.42024419', 'gathered_video_score2_overall_reward_mean': '9.58007812', 'gathered_video_score2_overall_reward_std': '1.35287011', 'gathered_video_score2_ta_reward_mean': '3.18969727', 'gathered_video_score2_ta_reward_std': '0.59152037', 'gathered_video_score2_vq_reward_mean': '3.09912109', 'gathered_video_score2_vq_reward_std': '0.60058492', 'gathered_ori_avg_reward_mean': '9.57812500', 'gathered_ori_avg_reward_std': '1.34968567'} +2026-06-26 18:49:15.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 18:49:15.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_0/index_6620_seed_1567_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 18:50:35.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:35.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:51:45.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_0/index_6620_seed_1568_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:52:04.201 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 18:52:05.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.77s (1.88min) +2026-06-26 18:53:58.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.577 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 208 train_steps 4320 {'step': '208/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00475368', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.76089478', 'grpo_loop_time': '422.14660645', 'ref_model_time': '112.77502441', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86474419', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.06005859', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.368e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-09', 'kl_loss': '0.00796290', 'kl_moving': '0.00796290', 'kl_stepwise': '-1.870e-08', 'gathered_avg_reward_mean': '9.06433105', 'gathered_avg_reward_std': '1.25697339', 'gathered_video_score2_mq_reward_mean': '3.12805176', 'gathered_video_score2_mq_reward_std': '0.37679526', 'gathered_video_score2_overall_reward_mean': '9.06005859', 'gathered_video_score2_overall_reward_std': '1.25586152', 'gathered_video_score2_ta_reward_mean': '2.91748047', 'gathered_video_score2_ta_reward_std': '0.79149395', 'gathered_video_score2_vq_reward_mean': '3.01879883', 'gathered_video_score2_vq_reward_std': '0.48166162', 'gathered_ori_avg_reward_mean': '9.06433105', 'gathered_ori_avg_reward_std': '1.25697351'} +2026-06-26 19:03:01.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 19:03:01.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:01.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:08.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_0/index_3393_seed_27727_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 19:04:22.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:22.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:35.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_0/index_3393_seed_27728_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:05:54.852 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 19:05:55.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.262 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 209 train_steps 4360 {'step': '209/10000', 'learning_rate': '0.00001000', 'loss': '1.118e-08', 'grad_norm': '0.00500380', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.75531006', 'grpo_loop_time': '418.84249878', 'ref_model_time': '113.15050507', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86993039', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.17333984', 'token_credit_video_advantage_mean': '1.164e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.438e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.118e-08', 'kl_loss': '0.00857031', 'kl_moving': '0.00857031', 'kl_stepwise': '-1.992e-08', 'gathered_avg_reward_mean': '9.17358398', 'gathered_avg_reward_std': '0.96576887', 'gathered_video_score2_mq_reward_mean': '3.15771484', 'gathered_video_score2_mq_reward_std': '0.27102336', 'gathered_video_score2_overall_reward_mean': '9.17333984', 'gathered_video_score2_overall_reward_std': '0.96404368', 'gathered_video_score2_ta_reward_mean': '3.06616211', 'gathered_video_score2_ta_reward_std': '0.49836266', 'gathered_video_score2_vq_reward_mean': '2.94970703', 'gathered_video_score2_vq_reward_std': '0.50535363', 'gathered_ori_avg_reward_mean': '9.17358398', 'gathered_ori_avg_reward_std': '0.96576893'} +2026-06-26 19:16:46.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 19:16:46.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:52.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_0/index_8829_seed_14580_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 19:18:12.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:18:12.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:20.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_0/index_8829_seed_14581_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:19:39.725 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 19:19:40.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:30:31.667 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 210 train_steps 4400 {'step': '210/10000', 'learning_rate': '0.00001000', 'loss': '-1.824e-08', 'grad_norm': '0.00546107', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.97140503', 'grpo_loop_time': '419.64215088', 'ref_model_time': '112.93460083', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85862160', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20898438', 'token_credit_video_advantage_mean': '8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.490e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.825e-08', 'kl_loss': '0.00885286', 'kl_moving': '0.00885287', 'kl_stepwise': '-1.994e-08', 'gathered_avg_reward_mean': '9.20605469', 'gathered_avg_reward_std': '1.09253681', 'gathered_video_score2_mq_reward_mean': '3.14709473', 'gathered_video_score2_mq_reward_std': '0.42059484', 'gathered_video_score2_overall_reward_mean': '9.20898438', 'gathered_video_score2_overall_reward_std': '1.09238017', 'gathered_video_score2_ta_reward_mean': '3.11181641', 'gathered_video_score2_ta_reward_std': '0.38040558', 'gathered_video_score2_vq_reward_mean': '2.94714355', 'gathered_video_score2_vq_reward_std': '0.57389772', 'gathered_ori_avg_reward_mean': '9.20605469', 'gathered_ori_avg_reward_std': '1.09253693'} +2026-06-26 19:31:06.988 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 19:31:09.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:17.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_0/index_1911_seed_6899_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 19:32:36.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:36.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:33:43.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_0/index_1911_seed_6900_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:34:06.233 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 19:34:07.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.772 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 211 train_steps 4440 {'step': '211/10000', 'learning_rate': '0.00001000', 'loss': '4.191e-09', 'grad_norm': '0.00561919', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.55145264', 'grpo_loop_time': '422.05102539', 'ref_model_time': '112.12528229', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85662168', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.73583984', 'token_credit_video_advantage_mean': '-4.889e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.418e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.191e-09', 'kl_loss': '0.00767428', 'kl_moving': '0.00767428', 'kl_stepwise': '-1.921e-08', 'gathered_avg_reward_mean': '8.73254395', 'gathered_avg_reward_std': '1.36829877', 'gathered_video_score2_mq_reward_mean': '3.03540039', 'gathered_video_score2_mq_reward_std': '0.41147706', 'gathered_video_score2_overall_reward_mean': '8.73583984', 'gathered_video_score2_overall_reward_std': '1.36657274', 'gathered_video_score2_ta_reward_mean': '2.91638184', 'gathered_video_score2_ta_reward_std': '0.62544036', 'gathered_video_score2_vq_reward_mean': '2.78076172', 'gathered_video_score2_vq_reward_std': '0.57920754', 'gathered_ori_avg_reward_mean': '8.73254395', 'gathered_ori_avg_reward_std': '1.36829877'} +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 19:45:03.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:09.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_0/index_1325_seed_9876_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 19:46:24.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:24.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:32.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_0/index_1325_seed_9877_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:47:51.825 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 19:47:52.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.670 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 212 train_steps 4480 {'step': '212/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00554872', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.66912842', 'grpo_loop_time': '420.66912842', 'ref_model_time': '112.18770599', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85989439', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.41162109', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.401e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.211e-08', 'kl_loss': '0.00706191', 'kl_moving': '0.00706191', 'kl_stepwise': '-1.958e-08', 'gathered_avg_reward_mean': '9.41333008', 'gathered_avg_reward_std': '0.81545097', 'gathered_video_score2_mq_reward_mean': '3.18627930', 'gathered_video_score2_mq_reward_std': '0.28475872', 'gathered_video_score2_overall_reward_mean': '9.41162109', 'gathered_video_score2_overall_reward_std': '0.81457359', 'gathered_video_score2_ta_reward_mean': '3.29174805', 'gathered_video_score2_ta_reward_std': '0.56805259', 'gathered_video_score2_vq_reward_mean': '2.93530273', 'gathered_video_score2_vq_reward_std': '0.40737399', 'gathered_ori_avg_reward_mean': '9.41333008', 'gathered_ori_avg_reward_std': '0.81545097'} +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 19:58:45.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:45.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_0/index_3774_seed_17673_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 20:00:05.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:00:05.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:17.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_0/index_3774_seed_17674_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:01:37.189 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 20:01:38.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.174 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 213 train_steps 4520 {'step': '213/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00555147', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.72415161', 'grpo_loop_time': '418.68576050', 'ref_model_time': '112.65434265', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85969591', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.92822266', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.497e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00773026', 'kl_moving': '0.00773026', 'kl_stepwise': '-1.914e-08', 'gathered_avg_reward_mean': '8.92358398', 'gathered_avg_reward_std': '1.21013749', 'gathered_video_score2_mq_reward_mean': '3.01525879', 'gathered_video_score2_mq_reward_std': '0.36201483', 'gathered_video_score2_overall_reward_mean': '8.92822266', 'gathered_video_score2_overall_reward_std': '1.21162629', 'gathered_video_score2_ta_reward_mean': '3.15356445', 'gathered_video_score2_ta_reward_std': '0.64061642', 'gathered_video_score2_vq_reward_mean': '2.75476074', 'gathered_video_score2_vq_reward_std': '0.59501857', 'gathered_ori_avg_reward_mean': '8.92358398', 'gathered_ori_avg_reward_std': '1.21013749'} +2026-06-26 20:12:28.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 20:12:28.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:34.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_0/index_8651_seed_3481_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 20:13:50.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:50.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:14:59.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_0/index_8651_seed_3482_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 20:15:21.128 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:15:21.280 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 20:15:22.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.038 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 214 train_steps 4560 {'step': '214/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00493264', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.89401245', 'grpo_loop_time': '418.76687622', 'ref_model_time': '112.15667725', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88122046', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.96191406', 'token_credit_video_advantage_mean': '-9.313e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.458e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00744360', 'kl_moving': '0.00744360', 'kl_stepwise': '-1.965e-08', 'gathered_avg_reward_mean': '8.96801758', 'gathered_avg_reward_std': '0.92231524', 'gathered_video_score2_mq_reward_mean': '3.15966797', 'gathered_video_score2_mq_reward_std': '0.27203140', 'gathered_video_score2_overall_reward_mean': '8.96191406', 'gathered_video_score2_overall_reward_std': '0.91919744', 'gathered_video_score2_ta_reward_mean': '2.84436035', 'gathered_video_score2_ta_reward_std': '0.57824260', 'gathered_video_score2_vq_reward_mean': '2.96398926', 'gathered_video_score2_vq_reward_std': '0.44325313', 'gathered_ori_avg_reward_mean': '8.96801758', 'gathered_ori_avg_reward_std': '0.92231530'} +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 20:26:14.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:20.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_0/index_3140_seed_9372_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 20:27:33.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:33.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:28:48.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_0/index_3140_seed_9373_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:29:11.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 20:29:12.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.790 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 215 train_steps 4600 {'step': '215/10000', 'learning_rate': '0.00001000', 'loss': '-8.382e-09', 'grad_norm': '0.00557954', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '293.44326782', 'grpo_loop_time': '419.22369385', 'ref_model_time': '114.06433868', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88318652', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.78369141', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.451e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.00799756', 'kl_moving': '0.00799756', 'kl_stepwise': '-1.988e-08', 'gathered_avg_reward_mean': '8.78344727', 'gathered_avg_reward_std': '1.32942653', 'gathered_video_score2_mq_reward_mean': '3.08728027', 'gathered_video_score2_mq_reward_std': '0.44576320', 'gathered_video_score2_overall_reward_mean': '8.78369141', 'gathered_video_score2_overall_reward_std': '1.33145607', 'gathered_video_score2_ta_reward_mean': '2.81750488', 'gathered_video_score2_ta_reward_std': '0.74413085', 'gathered_video_score2_vq_reward_mean': '2.87866211', 'gathered_video_score2_vq_reward_std': '0.57563531', 'gathered_ori_avg_reward_mean': '8.78344727', 'gathered_ori_avg_reward_std': '1.32942653'} +2026-06-26 20:40:06.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 20:40:06.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:06.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_0/index_9465_seed_28033_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 20:41:30.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:30.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:42:39.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_0/index_9465_seed_28034_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 20:43:00.483 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:43:00.611 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 20:43:01.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.670 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 216 train_steps 4640 {'step': '216/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00640949', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.22436523', 'grpo_loop_time': '418.64801025', 'ref_model_time': '113.70820618', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84389442', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.85107422', 'token_credit_video_advantage_mean': '-9.779e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-09', 'kl_loss': '0.00879006', 'kl_moving': '0.00879007', 'kl_stepwise': '-2.013e-08', 'gathered_avg_reward_mean': '8.85070801', 'gathered_avg_reward_std': '0.90308022', 'gathered_video_score2_mq_reward_mean': '3.13989258', 'gathered_video_score2_mq_reward_std': '0.25930896', 'gathered_video_score2_overall_reward_mean': '8.85107422', 'gathered_video_score2_overall_reward_std': '0.90280432', 'gathered_video_score2_ta_reward_mean': '2.83740234', 'gathered_video_score2_ta_reward_std': '0.62657100', 'gathered_video_score2_vq_reward_mean': '2.87341309', 'gathered_video_score2_vq_reward_std': '0.50749534', 'gathered_ori_avg_reward_mean': '8.85070801', 'gathered_ori_avg_reward_std': '0.90308022'} +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 20:53:52.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_0/index_6284_seed_23811_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 20:55:14.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:55:14.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:25.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_0/index_6284_seed_23812_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 20:56:53.176 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:56:53.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 20:56:54.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.465 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 217 train_steps 4680 {'step': '217/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00721866', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '295.14035034', 'grpo_loop_time': '418.46960449', 'ref_model_time': '112.90253448', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84625989', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.41162109', 'token_credit_video_advantage_mean': '-1.304e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.423e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00863620', 'kl_moving': '0.00863620', 'kl_stepwise': '-1.894e-08', 'gathered_avg_reward_mean': '8.40808105', 'gathered_avg_reward_std': '1.33562040', 'gathered_video_score2_mq_reward_mean': '2.94628906', 'gathered_video_score2_mq_reward_std': '0.38317728', 'gathered_video_score2_overall_reward_mean': '8.41162109', 'gathered_video_score2_overall_reward_std': '1.33683443', 'gathered_video_score2_ta_reward_mean': '2.76879883', 'gathered_video_score2_ta_reward_std': '0.69472349', 'gathered_video_score2_vq_reward_mean': '2.69299316', 'gathered_video_score2_vq_reward_std': '0.56842095', 'gathered_ori_avg_reward_mean': '8.40808105', 'gathered_ori_avg_reward_std': '1.33562028'} +2026-06-26 21:07:44.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 21:07:44.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:50.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_0/index_9434_seed_29584_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 21:09:06.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:09:06.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:18.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_0/index_9434_seed_29585_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 21:10:41.549 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:10:41.713 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 21:10:42.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.119 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 218 train_steps 4720 {'step': '218/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00499642', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.05953979', 'grpo_loop_time': '417.65637207', 'ref_model_time': '112.15206146', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85260755', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.04296875', 'token_credit_video_advantage_mean': '-1.024e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.373e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00871049', 'kl_moving': '0.00871049', 'kl_stepwise': '-1.881e-08', 'gathered_avg_reward_mean': '9.04528809', 'gathered_avg_reward_std': '1.23778224', 'gathered_video_score2_mq_reward_mean': '3.15307617', 'gathered_video_score2_mq_reward_std': '0.39332253', 'gathered_video_score2_overall_reward_mean': '9.04296875', 'gathered_video_score2_overall_reward_std': '1.24107528', 'gathered_video_score2_ta_reward_mean': '3.11535645', 'gathered_video_score2_ta_reward_std': '0.57517087', 'gathered_video_score2_vq_reward_mean': '2.77685547', 'gathered_video_score2_vq_reward_std': '0.52636784', 'gathered_ori_avg_reward_mean': '9.04528809', 'gathered_ori_avg_reward_std': '1.23778224'} +2026-06-26 21:21:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 21:21:31.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:37.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_0/index_5926_seed_8388_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 21:22:53.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:53.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:02.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_0/index_5926_seed_8389_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 21:24:25.878 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:24:26.043 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 21:24:26.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.622 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 219 train_steps 4760 {'step': '219/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00595687', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.47308350', 'grpo_loop_time': '417.21923828', 'ref_model_time': '112.02503204', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87558502', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03076172', 'token_credit_video_advantage_mean': '1.024e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.359e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00924051', 'kl_moving': '0.00924051', 'kl_stepwise': '-1.815e-08', 'gathered_avg_reward_mean': '9.03088379', 'gathered_avg_reward_std': '1.39671278', 'gathered_video_score2_mq_reward_mean': '3.14270020', 'gathered_video_score2_mq_reward_std': '0.43639615', 'gathered_video_score2_overall_reward_mean': '9.03076172', 'gathered_video_score2_overall_reward_std': '1.39845383', 'gathered_video_score2_ta_reward_mean': '3.10498047', 'gathered_video_score2_ta_reward_std': '0.63825345', 'gathered_video_score2_vq_reward_mean': '2.78320312', 'gathered_video_score2_vq_reward_std': '0.58257824', 'gathered_ori_avg_reward_mean': '9.03088379', 'gathered_ori_avg_reward_std': '1.39671278'} +2026-06-26 21:35:16.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 21:35:16.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:16.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_0/index_4438_seed_5525_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 21:36:37.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:37.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:37:49.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_0/index_4438_seed_5526_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:38:10.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 21:38:11.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:06.104 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 220 train_steps 4800 {'step': '220/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00639421', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.78479004', 'grpo_loop_time': '420.34304810', 'ref_model_time': '113.48253632', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86255485', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.42333984', 'token_credit_video_advantage_mean': '5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.448e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00883790', 'kl_moving': '0.00883790', 'kl_stepwise': '-1.935e-08', 'gathered_avg_reward_mean': '9.42175293', 'gathered_avg_reward_std': '1.37662113', 'gathered_video_score2_mq_reward_mean': '3.28247070', 'gathered_video_score2_mq_reward_std': '0.45299286', 'gathered_video_score2_overall_reward_mean': '9.42333984', 'gathered_video_score2_overall_reward_std': '1.37863433', 'gathered_video_score2_ta_reward_mean': '3.02819824', 'gathered_video_score2_ta_reward_std': '0.63832480', 'gathered_video_score2_vq_reward_mean': '3.11108398', 'gathered_video_score2_vq_reward_std': '0.52746034', 'gathered_ori_avg_reward_mean': '9.42175293', 'gathered_ori_avg_reward_std': '1.37662101'} +2026-06-26 21:49:41.510 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 21:49:43.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:43.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:51.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_0/index_2546_seed_31193_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 21:51:05.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:51:05.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:15.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_0/index_2546_seed_31194_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 21:52:37.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:52:37.861 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 21:52:39.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.559 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 221 train_steps 4840 {'step': '221/10000', 'learning_rate': '0.00001000', 'loss': '2.049e-08', 'grad_norm': '0.00418868', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.80499268', 'grpo_loop_time': '417.40380859', 'ref_model_time': '111.88430786', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88637006', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23046875', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.049e-08', 'kl_loss': '0.00730780', 'kl_moving': '0.00730780', 'kl_stepwise': '-2.008e-08', 'gathered_avg_reward_mean': '9.23071289', 'gathered_avg_reward_std': '1.20685351', 'gathered_video_score2_mq_reward_mean': '3.20361328', 'gathered_video_score2_mq_reward_std': '0.35291588', 'gathered_video_score2_overall_reward_mean': '9.23046875', 'gathered_video_score2_overall_reward_std': '1.20711970', 'gathered_video_score2_ta_reward_mean': '3.02160645', 'gathered_video_score2_ta_reward_std': '0.69381142', 'gathered_video_score2_vq_reward_mean': '3.00549316', 'gathered_video_score2_vq_reward_std': '0.58463508', 'gathered_ori_avg_reward_mean': '9.23071289', 'gathered_ori_avg_reward_std': '1.20685351'} +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 22:03:26.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_0/index_8306_seed_11615_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 22:04:48.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:48.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:05:56.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_0/index_8306_seed_11616_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:06:17.099 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 22:06:18.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:06.945 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 222 train_steps 4880 {'step': '222/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00569035', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.16467285', 'grpo_loop_time': '417.28683472', 'ref_model_time': '113.08641815', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86607194', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.49804688', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.404e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00751963', 'kl_moving': '0.00751963', 'kl_stepwise': '-1.911e-08', 'gathered_avg_reward_mean': '9.50170898', 'gathered_avg_reward_std': '0.95900869', 'gathered_video_score2_mq_reward_mean': '3.26123047', 'gathered_video_score2_mq_reward_std': '0.34162909', 'gathered_video_score2_overall_reward_mean': '9.49804688', 'gathered_video_score2_overall_reward_std': '0.95478654', 'gathered_video_score2_ta_reward_mean': '3.23767090', 'gathered_video_score2_ta_reward_std': '0.61944216', 'gathered_video_score2_vq_reward_mean': '3.00280762', 'gathered_video_score2_vq_reward_std': '0.39693719', 'gathered_ori_avg_reward_mean': '9.50170898', 'gathered_ori_avg_reward_std': '0.95900869'} +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 22:17:07.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:13.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_0/index_6626_seed_32833_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 22:18:31.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:31.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:37.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_0/index_6626_seed_32834_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:19:58.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 22:19:59.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.116 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 223 train_steps 4920 {'step': '223/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00565304', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.44760132', 'grpo_loop_time': '417.90625000', 'ref_model_time': '112.99523163', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86134505', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.84521484', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.507e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00829065', 'kl_moving': '0.00829065', 'kl_stepwise': '-1.977e-08', 'gathered_avg_reward_mean': '8.84484863', 'gathered_avg_reward_std': '1.05140543', 'gathered_video_score2_mq_reward_mean': '3.13623047', 'gathered_video_score2_mq_reward_std': '0.30603006', 'gathered_video_score2_overall_reward_mean': '8.84521484', 'gathered_video_score2_overall_reward_std': '1.04842508', 'gathered_video_score2_ta_reward_mean': '2.77502441', 'gathered_video_score2_ta_reward_std': '0.67721522', 'gathered_video_score2_vq_reward_mean': '2.93359375', 'gathered_video_score2_vq_reward_std': '0.47350091', 'gathered_ori_avg_reward_mean': '8.84484863', 'gathered_ori_avg_reward_std': '1.05140543'} +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 22:30:49.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:55.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_0/index_8544_seed_7706_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 22:32:14.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:32:14.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:20.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_0/index_8544_seed_7707_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 22:33:42.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:33:42.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 22:33:43.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.268 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 224 train_steps 4960 {'step': '224/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00405702', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.62713623', 'grpo_loop_time': '417.09527588', 'ref_model_time': '112.36936188', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85221070', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20214844', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.467e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00729039', 'kl_moving': '0.00729039', 'kl_stepwise': '-1.944e-08', 'gathered_avg_reward_mean': '9.20019531', 'gathered_avg_reward_std': '0.74414253', 'gathered_video_score2_mq_reward_mean': '3.11230469', 'gathered_video_score2_mq_reward_std': '0.22968742', 'gathered_video_score2_overall_reward_mean': '9.20214844', 'gathered_video_score2_overall_reward_std': '0.74581695', 'gathered_video_score2_ta_reward_mean': '3.14282227', 'gathered_video_score2_ta_reward_std': '0.49523845', 'gathered_video_score2_vq_reward_mean': '2.94506836', 'gathered_video_score2_vq_reward_std': '0.37570116', 'gathered_ori_avg_reward_mean': '9.20019531', 'gathered_ori_avg_reward_std': '0.74414253'} +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 22:44:32.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:38.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_0/index_3437_seed_7978_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 22:45:57.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:57.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:04.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_0/index_3437_seed_7979_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:47:23.303 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 22:47:24.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:58:14.650 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 225 train_steps 5000 {'step': '225/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00655022', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.10952759', 'grpo_loop_time': '418.29705811', 'ref_model_time': '111.94641113', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85873985', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.15185547', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.434e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00849179', 'kl_moving': '0.00849179', 'kl_stepwise': '-1.960e-08', 'gathered_avg_reward_mean': '9.15649414', 'gathered_avg_reward_std': '1.13816166', 'gathered_video_score2_mq_reward_mean': '3.08471680', 'gathered_video_score2_mq_reward_std': '0.30700496', 'gathered_video_score2_overall_reward_mean': '9.15185547', 'gathered_video_score2_overall_reward_std': '1.13550150', 'gathered_video_score2_ta_reward_mean': '3.23120117', 'gathered_video_score2_ta_reward_std': '0.50170743', 'gathered_video_score2_vq_reward_mean': '2.84057617', 'gathered_video_score2_vq_reward_std': '0.60688108', 'gathered_ori_avg_reward_mean': '9.15649414', 'gathered_ori_avg_reward_std': '1.13816166'} +2026-06-26 22:59:15.716 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 225 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-225 +2026-06-26 22:59:15.844 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 225: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-225 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 22:59:16.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:22.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_0/index_5886_seed_33483_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 23:00:37.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:37.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:01:48.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_0/index_5886_seed_33484_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:02:08.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 23:02:09.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.412 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 226 train_steps 5040 {'step': '226/10000', 'learning_rate': '0.00001000', 'loss': '-3.260e-09', 'grad_norm': '0.00778707', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.62097168', 'grpo_loop_time': '420.79306030', 'ref_model_time': '112.41521454', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87011999', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20800781', 'token_credit_video_advantage_mean': '-2.095e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.404e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.260e-09', 'kl_loss': '0.00826454', 'kl_moving': '0.00826454', 'kl_stepwise': '-1.997e-08', 'gathered_avg_reward_mean': '9.20983887', 'gathered_avg_reward_std': '1.13361275', 'gathered_video_score2_mq_reward_mean': '3.22094727', 'gathered_video_score2_mq_reward_std': '0.33758980', 'gathered_video_score2_overall_reward_mean': '9.20800781', 'gathered_video_score2_overall_reward_std': '1.13238430', 'gathered_video_score2_ta_reward_mean': '3.08081055', 'gathered_video_score2_ta_reward_std': '0.63635051', 'gathered_video_score2_vq_reward_mean': '2.90808105', 'gathered_video_score2_vq_reward_std': '0.50130433', 'gathered_ori_avg_reward_mean': '9.20983887', 'gathered_ori_avg_reward_std': '1.13361275'} +2026-06-26 23:13:02.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 23:13:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_0/index_2872_seed_21573_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 23:14:23.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:23.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:32.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_0/index_2872_seed_21574_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:15:52.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 23:15:53.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.251 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 227 train_steps 5080 {'step': '227/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.00442391', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.88449097', 'grpo_loop_time': '418.77929688', 'ref_model_time': '112.46071625', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86832052', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.41210938', 'token_credit_video_advantage_mean': '-4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.440e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.794e-09', 'kl_loss': '0.00746949', 'kl_moving': '0.00746950', 'kl_stepwise': '-1.935e-08', 'gathered_avg_reward_mean': '9.41247559', 'gathered_avg_reward_std': '1.20043659', 'gathered_video_score2_mq_reward_mean': '3.16992188', 'gathered_video_score2_mq_reward_std': '0.41735223', 'gathered_video_score2_overall_reward_mean': '9.41210938', 'gathered_video_score2_overall_reward_std': '1.19872713', 'gathered_video_score2_ta_reward_mean': '3.17846680', 'gathered_video_score2_ta_reward_std': '0.62242395', 'gathered_video_score2_vq_reward_mean': '3.06408691', 'gathered_video_score2_vq_reward_std': '0.44728521', 'gathered_ori_avg_reward_mean': '9.41247559', 'gathered_ori_avg_reward_std': '1.20043659'} +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 23:26:44.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:50.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_0/index_9926_seed_3387_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 23:28:10.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:28:10.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:16.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_0/index_9926_seed_3388_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 23:29:38.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:29:38.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 23:29:39.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.499 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 228 train_steps 5120 {'step': '228/10000', 'learning_rate': '0.00001000', 'loss': '-9.872e-09', 'grad_norm': '0.00532305', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.89916992', 'grpo_loop_time': '417.32931519', 'ref_model_time': '111.83306122', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86985791', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.93896484', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.410e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.015e-08', 'kl_loss': '0.00860347', 'kl_moving': '0.00860347', 'kl_stepwise': '-1.942e-08', 'gathered_avg_reward_mean': '8.93884277', 'gathered_avg_reward_std': '0.97077924', 'gathered_video_score2_mq_reward_mean': '3.07958984', 'gathered_video_score2_mq_reward_std': '0.34422374', 'gathered_video_score2_overall_reward_mean': '8.93896484', 'gathered_video_score2_overall_reward_std': '0.97594404', 'gathered_video_score2_ta_reward_mean': '2.97961426', 'gathered_video_score2_ta_reward_std': '0.60541302', 'gathered_video_score2_vq_reward_mean': '2.87963867', 'gathered_video_score2_vq_reward_std': '0.44087601', 'gathered_ori_avg_reward_mean': '8.93884277', 'gathered_ori_avg_reward_std': '0.97077930'} +2026-06-26 23:40:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 23:40:27.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:33.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_0/index_4166_seed_4476_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 23:41:50.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:50.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:01.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_0/index_4166_seed_4477_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:43:25.429 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 23:43:26.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.881 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 229 train_steps 5160 {'step': '229/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00710398', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.04846191', 'grpo_loop_time': '416.70562744', 'ref_model_time': '112.56537628', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84750265', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.97412109', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.375e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.211e-08', 'kl_loss': '0.00956709', 'kl_moving': '0.00956709', 'kl_stepwise': '-1.953e-08', 'gathered_avg_reward_mean': '8.97436523', 'gathered_avg_reward_std': '1.16495883', 'gathered_video_score2_mq_reward_mean': '3.10070801', 'gathered_video_score2_mq_reward_std': '0.33031818', 'gathered_video_score2_overall_reward_mean': '8.97412109', 'gathered_video_score2_overall_reward_std': '1.16455472', 'gathered_video_score2_ta_reward_mean': '2.98571777', 'gathered_video_score2_ta_reward_std': '0.58196992', 'gathered_video_score2_vq_reward_mean': '2.88793945', 'gathered_video_score2_vq_reward_std': '0.55933458', 'gathered_ori_avg_reward_mean': '8.97436523', 'gathered_ori_avg_reward_std': '1.16495883'} +2026-06-26 23:54:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-26 23:54:16.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_0/index_9202_seed_27401_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-26 23:55:37.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:37.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:56:47.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_0/index_9202_seed_27402_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:57:08.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-26 23:57:09.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:02.544 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 230 train_steps 5200 {'step': '230/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00723809', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.00085449', 'grpo_loop_time': '421.12734985', 'ref_model_time': '112.35380554', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85585129', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.88037109', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.581e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-09', 'kl_loss': '0.00947866', 'kl_moving': '0.00947867', 'kl_stepwise': '-1.992e-08', 'gathered_avg_reward_mean': '8.88317871', 'gathered_avg_reward_std': '1.36617041', 'gathered_video_score2_mq_reward_mean': '3.08081055', 'gathered_video_score2_mq_reward_std': '0.45140901', 'gathered_video_score2_overall_reward_mean': '8.88037109', 'gathered_video_score2_overall_reward_std': '1.36774051', 'gathered_video_score2_ta_reward_mean': '2.94470215', 'gathered_video_score2_ta_reward_std': '0.77006310', 'gathered_video_score2_vq_reward_mean': '2.85766602', 'gathered_video_score2_vq_reward_std': '0.56070930', 'gathered_ori_avg_reward_mean': '8.88317871', 'gathered_ori_avg_reward_std': '1.36617041'} +2026-06-27 00:08:38.148 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 00:08:40.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:48.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_0/index_8496_seed_150_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 00:10:03.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:10:03.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:12.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_0/index_8496_seed_151_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:11:31.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 00:11:32.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.108 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 231 train_steps 5240 {'step': '231/10000', 'learning_rate': '0.00001000', 'loss': '6.519e-09', 'grad_norm': '0.00469851', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.64404297', 'grpo_loop_time': '417.71401978', 'ref_model_time': '111.31204987', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86331773', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.50048828', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00744123', 'kl_moving': '0.00744123', 'kl_stepwise': '-1.978e-08', 'gathered_avg_reward_mean': '9.50280762', 'gathered_avg_reward_std': '1.54952526', 'gathered_video_score2_mq_reward_mean': '3.20593262', 'gathered_video_score2_mq_reward_std': '0.50260317', 'gathered_video_score2_overall_reward_mean': '9.50048828', 'gathered_video_score2_overall_reward_std': '1.55068278', 'gathered_video_score2_ta_reward_mean': '3.25793457', 'gathered_video_score2_ta_reward_std': '0.61281067', 'gathered_video_score2_vq_reward_mean': '3.03894043', 'gathered_video_score2_vq_reward_std': '0.64364803', 'gathered_ori_avg_reward_mean': '9.50280762', 'gathered_ori_avg_reward_std': '1.54952526'} +2026-06-27 00:22:20.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 00:22:20.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:26.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_0/index_5465_seed_19254_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 00:23:44.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:44.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:24:54.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_0/index_5465_seed_19255_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 00:25:14.490 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:25:14.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 00:25:15.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.308 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 232 train_steps 5280 {'step': '232/10000', 'learning_rate': '0.00001000', 'loss': '9.779e-09', 'grad_norm': '0.00585551', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.16894531', 'grpo_loop_time': '418.70883179', 'ref_model_time': '113.18080139', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88165122', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.44726562', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.503e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.054e-09', 'kl_loss': '0.00813269', 'kl_moving': '0.00813269', 'kl_stepwise': '-2.020e-08', 'gathered_avg_reward_mean': '9.44897461', 'gathered_avg_reward_std': '0.80451792', 'gathered_video_score2_mq_reward_mean': '3.21435547', 'gathered_video_score2_mq_reward_std': '0.28741440', 'gathered_video_score2_overall_reward_mean': '9.44726562', 'gathered_video_score2_overall_reward_std': '0.80610603', 'gathered_video_score2_ta_reward_mean': '3.22998047', 'gathered_video_score2_ta_reward_std': '0.47105122', 'gathered_video_score2_vq_reward_mean': '3.00463867', 'gathered_video_score2_vq_reward_std': '0.39676502', 'gathered_ori_avg_reward_mean': '9.44897461', 'gathered_ori_avg_reward_std': '0.80451798'} +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 00:36:06.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:12.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_0/index_1410_seed_31397_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 00:37:26.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:26.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:38.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_0/index_1410_seed_31398_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:38:58.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 00:38:59.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.200 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 233 train_steps 5320 {'step': '233/10000', 'learning_rate': '0.00001000', 'loss': '8.848e-09', 'grad_norm': '0.00655230', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.57241821', 'grpo_loop_time': '418.17425537', 'ref_model_time': '113.02309418', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87293500', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.79931641', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.416e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.848e-09', 'kl_loss': '0.00850717', 'kl_moving': '0.00850718', 'kl_stepwise': '-1.981e-08', 'gathered_avg_reward_mean': '8.79724121', 'gathered_avg_reward_std': '0.96571261', 'gathered_video_score2_mq_reward_mean': '3.07666016', 'gathered_video_score2_mq_reward_std': '0.29169932', 'gathered_video_score2_overall_reward_mean': '8.79931641', 'gathered_video_score2_overall_reward_std': '0.96060085', 'gathered_video_score2_ta_reward_mean': '2.89367676', 'gathered_video_score2_ta_reward_std': '0.50000691', 'gathered_video_score2_vq_reward_mean': '2.82690430', 'gathered_video_score2_vq_reward_std': '0.47499546', 'gathered_ori_avg_reward_mean': '8.79724121', 'gathered_ori_avg_reward_std': '0.96571261'} +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 00:49:49.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:55.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_0/index_6838_seed_31328_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 00:51:11.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:51:11.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:20.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_0/index_6838_seed_31329_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 00:52:43.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:52:44.107 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 00:52:44.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.062 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 234 train_steps 5360 {'step': '234/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.01252258', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.90710449', 'grpo_loop_time': '418.80496216', 'ref_model_time': '111.47528839', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88482910', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.75097656', 'token_credit_video_advantage_mean': '-8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.444e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00797129', 'kl_moving': '0.00797129', 'kl_stepwise': '-1.901e-08', 'gathered_avg_reward_mean': '9.75195312', 'gathered_avg_reward_std': '0.88011420', 'gathered_video_score2_mq_reward_mean': '3.33032227', 'gathered_video_score2_mq_reward_std': '0.32342792', 'gathered_video_score2_overall_reward_mean': '9.75097656', 'gathered_video_score2_overall_reward_std': '0.88474566', 'gathered_video_score2_ta_reward_mean': '3.32861328', 'gathered_video_score2_ta_reward_std': '0.43352309', 'gathered_video_score2_vq_reward_mean': '3.09301758', 'gathered_video_score2_vq_reward_std': '0.48553470', 'gathered_ori_avg_reward_mean': '9.75195312', 'gathered_ori_avg_reward_std': '0.88011420'} +2026-06-27 01:03:35.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 01:03:35.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:41.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_0/index_8760_seed_24004_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 01:04:57.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:57.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:06.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_0/index_8760_seed_24005_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 01:06:27.584 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:06:27.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 01:06:28.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.698 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 235 train_steps 5400 {'step': '235/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00461598', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.07864380', 'grpo_loop_time': '419.06304932', 'ref_model_time': '111.82465363', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84582627', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.97509766', 'token_credit_video_advantage_mean': '7.916e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.430e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00763054', 'kl_moving': '0.00763054', 'kl_stepwise': '-1.856e-08', 'gathered_avg_reward_mean': '8.97326660', 'gathered_avg_reward_std': '1.00719798', 'gathered_video_score2_mq_reward_mean': '3.07080078', 'gathered_video_score2_mq_reward_std': '0.32726321', 'gathered_video_score2_overall_reward_mean': '8.97509766', 'gathered_video_score2_overall_reward_std': '1.01065767', 'gathered_video_score2_ta_reward_mean': '2.90625000', 'gathered_video_score2_ta_reward_std': '0.49987218', 'gathered_video_score2_vq_reward_mean': '2.99621582', 'gathered_video_score2_vq_reward_std': '0.47457519', 'gathered_ori_avg_reward_mean': '8.97326660', 'gathered_ori_avg_reward_std': '1.00719798'} +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 01:17:19.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_0/index_5260_seed_31378_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 01:18:39.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:39.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:19:52.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_0/index_5260_seed_31379_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 01:20:14.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:20:14.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 01:20:15.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.779 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 236 train_steps 5440 {'step': '236/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.01034661', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.00277710', 'grpo_loop_time': '418.77737427', 'ref_model_time': '113.56268311', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84728503', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.06884766', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.473e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.794e-09', 'kl_loss': '0.00877299', 'kl_moving': '0.00877300', 'kl_stepwise': '-1.893e-08', 'gathered_avg_reward_mean': '9.07128906', 'gathered_avg_reward_std': '0.90581369', 'gathered_video_score2_mq_reward_mean': '3.05407715', 'gathered_video_score2_mq_reward_std': '0.24795191', 'gathered_video_score2_overall_reward_mean': '9.06884766', 'gathered_video_score2_overall_reward_std': '0.90586567', 'gathered_video_score2_ta_reward_mean': '3.20605469', 'gathered_video_score2_ta_reward_std': '0.64290082', 'gathered_video_score2_vq_reward_mean': '2.81115723', 'gathered_video_score2_vq_reward_std': '0.50820404', 'gathered_ori_avg_reward_mean': '9.07128906', 'gathered_ori_avg_reward_std': '0.90581369'} +2026-06-27 01:31:06.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 01:31:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:06.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:13.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_0/index_8451_seed_20629_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 01:32:30.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:30.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:39.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_0/index_8451_seed_20630_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:33:58.124 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 01:33:58.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.563 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 237 train_steps 5480 {'step': '237/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00653345', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.61199951', 'grpo_loop_time': '417.43121338', 'ref_model_time': '112.65893555', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87814659', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07080078', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.552e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00869427', 'kl_moving': '0.00869427', 'kl_stepwise': '-2.024e-08', 'gathered_avg_reward_mean': '9.07360840', 'gathered_avg_reward_std': '1.36578524', 'gathered_video_score2_mq_reward_mean': '3.11193848', 'gathered_video_score2_mq_reward_std': '0.49077693', 'gathered_video_score2_overall_reward_mean': '9.07080078', 'gathered_video_score2_overall_reward_std': '1.36330080', 'gathered_video_score2_ta_reward_mean': '3.21240234', 'gathered_video_score2_ta_reward_std': '0.58970612', 'gathered_video_score2_vq_reward_mean': '2.74926758', 'gathered_video_score2_vq_reward_std': '0.51058573', 'gathered_ori_avg_reward_mean': '9.07360840', 'gathered_ori_avg_reward_std': '1.36578524'} +2026-06-27 01:44:47.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 01:44:47.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_0/index_5609_seed_28072_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 01:46:11.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:46:11.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:19.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_0/index_5609_seed_28073_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 01:47:39.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:47:39.678 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 01:47:40.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.367 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 238 train_steps 5520 {'step': '238/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00618043', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.06625366', 'grpo_loop_time': '417.10015869', 'ref_model_time': '111.38613892', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84760475', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.19824219', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.408e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00886289', 'kl_moving': '0.00886290', 'kl_stepwise': '-1.981e-08', 'gathered_avg_reward_mean': '9.20104980', 'gathered_avg_reward_std': '0.91719884', 'gathered_video_score2_mq_reward_mean': '3.16577148', 'gathered_video_score2_mq_reward_std': '0.28793353', 'gathered_video_score2_overall_reward_mean': '9.19824219', 'gathered_video_score2_overall_reward_std': '0.91813028', 'gathered_video_score2_ta_reward_mean': '3.09265137', 'gathered_video_score2_ta_reward_std': '0.56777662', 'gathered_video_score2_vq_reward_mean': '2.94262695', 'gathered_video_score2_vq_reward_std': '0.36185092', 'gathered_ori_avg_reward_mean': '9.20104980', 'gathered_ori_avg_reward_std': '0.91719884'} +2026-06-27 01:58:27.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 01:58:27.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:33.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_0/index_4937_seed_20629_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 01:59:51.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:52.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:01.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_0/index_4937_seed_20630_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:01:22.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 02:01:23.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.608 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 239 train_steps 5560 {'step': '239/10000', 'learning_rate': '0.00001000', 'loss': '2.328e-09', 'grad_norm': '0.00663059', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.02819824', 'grpo_loop_time': '419.38842773', 'ref_model_time': '111.09824371', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86473644', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.39648438', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.390e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-09', 'kl_loss': '0.00835395', 'kl_moving': '0.00835396', 'kl_stepwise': '-1.862e-08', 'gathered_avg_reward_mean': '9.39562988', 'gathered_avg_reward_std': '0.96338701', 'gathered_video_score2_mq_reward_mean': '3.23168945', 'gathered_video_score2_mq_reward_std': '0.36221445', 'gathered_video_score2_overall_reward_mean': '9.39648438', 'gathered_video_score2_overall_reward_std': '0.96586126', 'gathered_video_score2_ta_reward_mean': '3.04162598', 'gathered_video_score2_ta_reward_std': '0.53226876', 'gathered_video_score2_vq_reward_mean': '3.12231445', 'gathered_video_score2_vq_reward_std': '0.46341822', 'gathered_ori_avg_reward_mean': '9.39562988', 'gathered_ori_avg_reward_std': '0.96338707'} +2026-06-27 02:12:14.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 02:12:14.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:21.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_0/index_6887_seed_19194_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 02:13:37.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:37.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:14:47.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_0/index_6887_seed_19195_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:15:08.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 02:15:09.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:02.002 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 240 train_steps 5600 {'step': '240/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00554966', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.10559082', 'grpo_loop_time': '420.63259888', 'ref_model_time': '112.24085999', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86480910', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.17285156', 'token_credit_video_advantage_mean': '-1.688e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.486e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00866822', 'kl_moving': '0.00866822', 'kl_stepwise': '-1.979e-08', 'gathered_avg_reward_mean': '9.17016602', 'gathered_avg_reward_std': '0.96542722', 'gathered_video_score2_mq_reward_mean': '3.15539551', 'gathered_video_score2_mq_reward_std': '0.35966179', 'gathered_video_score2_overall_reward_mean': '9.17285156', 'gathered_video_score2_overall_reward_std': '0.96442407', 'gathered_video_score2_ta_reward_mean': '2.96569824', 'gathered_video_score2_ta_reward_std': '0.54864496', 'gathered_video_score2_vq_reward_mean': '3.04907227', 'gathered_video_score2_vq_reward_std': '0.55080801', 'gathered_ori_avg_reward_mean': '9.17016602', 'gathered_ori_avg_reward_std': '0.96542722'} +2026-06-27 02:26:37.319 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 02:26:39.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_0/index_9944_seed_13142_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 02:28:00.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:28:00.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:13.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_0/index_9944_seed_13143_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:29:33.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 02:29:34.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.713 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 241 train_steps 5640 {'step': '241/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00535100', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.73669434', 'grpo_loop_time': '418.24539185', 'ref_model_time': '112.05078125', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85801524', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.70019531', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.433e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-09', 'kl_loss': '0.00750071', 'kl_moving': '0.00750072', 'kl_stepwise': '-1.949e-08', 'gathered_avg_reward_mean': '8.70190430', 'gathered_avg_reward_std': '1.13330507', 'gathered_video_score2_mq_reward_mean': '3.01586914', 'gathered_video_score2_mq_reward_std': '0.34322238', 'gathered_video_score2_overall_reward_mean': '8.70019531', 'gathered_video_score2_overall_reward_std': '1.13166320', 'gathered_video_score2_ta_reward_mean': '2.87255859', 'gathered_video_score2_ta_reward_std': '0.53425890', 'gathered_video_score2_vq_reward_mean': '2.81347656', 'gathered_video_score2_vq_reward_std': '0.51491708', 'gathered_ori_avg_reward_mean': '8.70190430', 'gathered_ori_avg_reward_std': '1.13330507'} +2026-06-27 02:40:22.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 02:40:23.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_0/index_2462_seed_16563_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 02:41:41.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:41.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:00.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_0/index_2462_seed_16564_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:43:24.620 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 02:43:25.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.395 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 242 train_steps 5680 {'step': '242/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00487122', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '295.15115356', 'grpo_loop_time': '417.94662476', 'ref_model_time': '111.86936951', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87908250', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24169922', 'token_credit_video_advantage_mean': '4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.449e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00779680', 'kl_moving': '0.00779680', 'kl_stepwise': '-1.908e-08', 'gathered_avg_reward_mean': '9.23986816', 'gathered_avg_reward_std': '1.25001419', 'gathered_video_score2_mq_reward_mean': '3.17871094', 'gathered_video_score2_mq_reward_std': '0.46335497', 'gathered_video_score2_overall_reward_mean': '9.24169922', 'gathered_video_score2_overall_reward_std': '1.24730229', 'gathered_video_score2_ta_reward_mean': '2.99133301', 'gathered_video_score2_ta_reward_std': '0.53534710', 'gathered_video_score2_vq_reward_mean': '3.06982422', 'gathered_video_score2_vq_reward_std': '0.48483130', 'gathered_ori_avg_reward_mean': '9.23986816', 'gathered_ori_avg_reward_std': '1.25001431'} +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 02:54:14.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:20.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_0/index_7728_seed_5274_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 02:55:34.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:34.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:56:47.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_0/index_7728_seed_5275_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:57:08.549 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 02:57:09.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.631 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 243 train_steps 5720 {'step': '243/10000', 'learning_rate': '0.00001000', 'loss': '2.328e-09', 'grad_norm': '0.00735833', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.25958252', 'grpo_loop_time': '420.52490234', 'ref_model_time': '111.55477905', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87398058', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07666016', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.453e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.122e-09', 'kl_loss': '0.00836060', 'kl_moving': '0.00836060', 'kl_stepwise': '-1.968e-08', 'gathered_avg_reward_mean': '9.07543945', 'gathered_avg_reward_std': '0.95104575', 'gathered_video_score2_mq_reward_mean': '3.10510254', 'gathered_video_score2_mq_reward_std': '0.31203863', 'gathered_video_score2_overall_reward_mean': '9.07666016', 'gathered_video_score2_overall_reward_std': '0.95173007', 'gathered_video_score2_ta_reward_mean': '3.05810547', 'gathered_video_score2_ta_reward_std': '0.54121101', 'gathered_video_score2_vq_reward_mean': '2.91223145', 'gathered_video_score2_vq_reward_std': '0.50192606', 'gathered_ori_avg_reward_mean': '9.07543945', 'gathered_ori_avg_reward_std': '0.95104581'} +2026-06-27 03:08:01.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 03:08:01.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_0/index_5308_seed_8721_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 03:09:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:25.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_0/index_5308_seed_8722_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 03:10:49.205 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:10:49.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 03:10:50.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.416 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 244 train_steps 5760 {'step': '244/10000', 'learning_rate': '0.00001000', 'loss': '-1.304e-08', 'grad_norm': '0.00638661', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '280.83187866', 'grpo_loop_time': '419.44775391', 'ref_model_time': '111.55767059', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86789733', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.38818359', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.304e-08', 'kl_loss': '0.00750029', 'kl_moving': '0.00750029', 'kl_stepwise': '-1.966e-08', 'gathered_avg_reward_mean': '9.39196777', 'gathered_avg_reward_std': '1.00320065', 'gathered_video_score2_mq_reward_mean': '3.24902344', 'gathered_video_score2_mq_reward_std': '0.35850221', 'gathered_video_score2_overall_reward_mean': '9.38818359', 'gathered_video_score2_overall_reward_std': '1.00254202', 'gathered_video_score2_ta_reward_mean': '3.07849121', 'gathered_video_score2_ta_reward_std': '0.63323349', 'gathered_video_score2_vq_reward_mean': '3.06445312', 'gathered_video_score2_vq_reward_std': '0.54715568', 'gathered_ori_avg_reward_mean': '9.39196777', 'gathered_ori_avg_reward_std': '1.00320077'} +2026-06-27 03:21:41.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 03:21:41.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:47.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_0/index_5288_seed_8132_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 03:23:02.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:23:02.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:15.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_0/index_5288_seed_8133_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:24:37.787 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 03:24:38.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.303 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 245 train_steps 5800 {'step': '245/10000', 'learning_rate': '0.00001000', 'loss': '3.260e-09', 'grad_norm': '0.00554827', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.22265625', 'grpo_loop_time': '418.06933594', 'ref_model_time': '111.26432800', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85192031', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.18212891', 'token_credit_video_advantage_mean': '2.561e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.378e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.260e-09', 'kl_loss': '0.00793100', 'kl_moving': '0.00793100', 'kl_stepwise': '-1.885e-08', 'gathered_avg_reward_mean': '9.18322754', 'gathered_avg_reward_std': '0.88795894', 'gathered_video_score2_mq_reward_mean': '3.11730957', 'gathered_video_score2_mq_reward_std': '0.27779552', 'gathered_video_score2_overall_reward_mean': '9.18212891', 'gathered_video_score2_overall_reward_std': '0.88986319', 'gathered_video_score2_ta_reward_mean': '3.01806641', 'gathered_video_score2_ta_reward_std': '0.66557568', 'gathered_video_score2_vq_reward_mean': '3.04785156', 'gathered_video_score2_vq_reward_std': '0.33305076', 'gathered_ori_avg_reward_mean': '9.18322754', 'gathered_ori_avg_reward_std': '0.88795894'} +2026-06-27 03:35:26.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 03:35:26.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:32.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_0/index_3878_seed_25920_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 03:36:52.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:52.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:03.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_0/index_3878_seed_25921_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 03:38:24.544 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:38:24.713 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 03:38:25.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.382 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 246 train_steps 5840 {'step': '246/10000', 'learning_rate': '0.00001000', 'loss': '-3.260e-09', 'grad_norm': '0.00498567', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.60134888', 'grpo_loop_time': '418.32183838', 'ref_model_time': '111.71736908', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87835258', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.45751953', 'token_credit_video_advantage_mean': '-7.916e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.435e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.260e-09', 'kl_loss': '0.00774281', 'kl_moving': '0.00774281', 'kl_stepwise': '-1.857e-08', 'gathered_avg_reward_mean': '9.45629883', 'gathered_avg_reward_std': '1.23542738', 'gathered_video_score2_mq_reward_mean': '3.29370117', 'gathered_video_score2_mq_reward_std': '0.45141429', 'gathered_video_score2_overall_reward_mean': '9.45751953', 'gathered_video_score2_overall_reward_std': '1.23152077', 'gathered_video_score2_ta_reward_mean': '3.05773926', 'gathered_video_score2_ta_reward_std': '0.66421419', 'gathered_video_score2_vq_reward_mean': '3.10485840', 'gathered_video_score2_vq_reward_std': '0.59953272', 'gathered_ori_avg_reward_mean': '9.45629883', 'gathered_ori_avg_reward_std': '1.23542738'} +2026-06-27 03:49:14.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 03:49:14.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:20.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_0/index_1422_seed_4519_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 03:50:36.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:36.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:51:45.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_0/index_1422_seed_4520_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:52:07.529 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 03:52:08.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.073 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 247 train_steps 5880 {'step': '247/10000', 'learning_rate': '0.00001000', 'loss': '2.701e-08', 'grad_norm': '0.00677049', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.63217163', 'grpo_loop_time': '421.58908081', 'ref_model_time': '111.66528320', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83599699', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.47998047', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.392e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.701e-08', 'kl_loss': '0.00811877', 'kl_moving': '0.00811877', 'kl_stepwise': '-1.877e-08', 'gathered_avg_reward_mean': '9.47485352', 'gathered_avg_reward_std': '0.85267437', 'gathered_video_score2_mq_reward_mean': '3.16809082', 'gathered_video_score2_mq_reward_std': '0.42910337', 'gathered_video_score2_overall_reward_mean': '9.47998047', 'gathered_video_score2_overall_reward_std': '0.85419905', 'gathered_video_score2_ta_reward_mean': '3.18981934', 'gathered_video_score2_ta_reward_std': '0.57917267', 'gathered_video_score2_vq_reward_mean': '3.11694336', 'gathered_video_score2_vq_reward_std': '0.43516564', 'gathered_ori_avg_reward_mean': '9.47485352', 'gathered_ori_avg_reward_std': '0.85267437'} +2026-06-27 04:03:03.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 04:03:03.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:09.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_0/index_9458_seed_866_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 04:04:25.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:25.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:34.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_0/index_9458_seed_867_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:05:54.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 04:05:55.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.868 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 248 train_steps 5920 {'step': '248/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00567002', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.60800171', 'grpo_loop_time': '420.27999878', 'ref_model_time': '114.03649902', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84836841', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.98535156', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.407e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00845405', 'kl_moving': '0.00845405', 'kl_stepwise': '-1.895e-08', 'gathered_avg_reward_mean': '8.98071289', 'gathered_avg_reward_std': '0.94662201', 'gathered_video_score2_mq_reward_mean': '3.10937500', 'gathered_video_score2_mq_reward_std': '0.26451689', 'gathered_video_score2_overall_reward_mean': '8.98535156', 'gathered_video_score2_overall_reward_std': '0.95139390', 'gathered_video_score2_ta_reward_mean': '2.97802734', 'gathered_video_score2_ta_reward_std': '0.52814406', 'gathered_video_score2_vq_reward_mean': '2.89331055', 'gathered_video_score2_vq_reward_std': '0.38633934', 'gathered_ori_avg_reward_mean': '8.98071289', 'gathered_ori_avg_reward_std': '0.94662207'} +2026-06-27 04:16:47.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 04:16:48.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_0/index_3484_seed_26788_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 04:18:10.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:18:10.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:20.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_0/index_3484_seed_26789_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:19:41.734 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 04:19:42.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.784 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 249 train_steps 5960 {'step': '249/10000', 'learning_rate': '0.00001000', 'loss': '-4.377e-09', 'grad_norm': '0.00746023', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.95138550', 'grpo_loop_time': '418.03390503', 'ref_model_time': '111.44685364', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85622990', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91357422', 'token_credit_video_advantage_mean': '-4.540e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.425e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.377e-09', 'kl_loss': '0.00901323', 'kl_moving': '0.00901323', 'kl_stepwise': '-1.941e-08', 'gathered_avg_reward_mean': '8.90905762', 'gathered_avg_reward_std': '1.25688744', 'gathered_video_score2_mq_reward_mean': '3.03515625', 'gathered_video_score2_mq_reward_std': '0.44042644', 'gathered_video_score2_overall_reward_mean': '8.91357422', 'gathered_video_score2_overall_reward_std': '1.25614631', 'gathered_video_score2_ta_reward_mean': '3.18078613', 'gathered_video_score2_ta_reward_std': '0.52940655', 'gathered_video_score2_vq_reward_mean': '2.69311523', 'gathered_video_score2_vq_reward_std': '0.54895401', 'gathered_ori_avg_reward_mean': '8.90905762', 'gathered_ori_avg_reward_std': '1.25688744'} +2026-06-27 04:30:30.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 04:30:30.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:30.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_0/index_4688_seed_3903_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 04:31:53.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:53.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:01.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_0/index_4688_seed_3904_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 04:33:21.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:33:21.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 04:33:22.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:44:14.215 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 250 train_steps 6000 {'step': '250/10000', 'learning_rate': '0.00001000', 'loss': '-1.397e-09', 'grad_norm': '0.00554850', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.13021851', 'grpo_loop_time': '419.18823242', 'ref_model_time': '111.67994690', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86707348', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.48437500', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.462e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.397e-09', 'kl_loss': '0.00876609', 'kl_moving': '0.00876609', 'kl_stepwise': '-1.997e-08', 'gathered_avg_reward_mean': '9.48388672', 'gathered_avg_reward_std': '0.73751599', 'gathered_video_score2_mq_reward_mean': '3.19824219', 'gathered_video_score2_mq_reward_std': '0.21652177', 'gathered_video_score2_overall_reward_mean': '9.48437500', 'gathered_video_score2_overall_reward_std': '0.74008751', 'gathered_video_score2_ta_reward_mean': '3.23437500', 'gathered_video_score2_ta_reward_std': '0.43766564', 'gathered_video_score2_vq_reward_mean': '3.05126953', 'gathered_video_score2_vq_reward_std': '0.43067595', 'gathered_ori_avg_reward_mean': '9.48388672', 'gathered_ori_avg_reward_std': '0.73751599'} +2026-06-27 04:45:18.111 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 250 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-250 +2026-06-27 04:45:18.238 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 250: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-250 +2026-06-27 04:45:21.084 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-225 +2026-06-27 04:45:56.569 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 04:45:58.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_0/index_4044_seed_18127_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 04:47:24.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:24.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:31.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_0/index_4044_seed_18128_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 04:48:55.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:48:56.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 04:48:56.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.497 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 251 train_steps 6040 {'step': '251/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00575334', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.53787231', 'grpo_loop_time': '418.48684692', 'ref_model_time': '112.27481842', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86404997', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.03125000', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.506e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.00793696', 'kl_moving': '0.00793696', 'kl_stepwise': '-1.990e-08', 'gathered_avg_reward_mean': '9.02941895', 'gathered_avg_reward_std': '1.17931533', 'gathered_video_score2_mq_reward_mean': '3.11437988', 'gathered_video_score2_mq_reward_std': '0.32776839', 'gathered_video_score2_overall_reward_mean': '9.03125000', 'gathered_video_score2_overall_reward_std': '1.18144524', 'gathered_video_score2_ta_reward_mean': '2.83764648', 'gathered_video_score2_ta_reward_std': '0.70817208', 'gathered_video_score2_vq_reward_mean': '3.07739258', 'gathered_video_score2_vq_reward_std': '0.42252040', 'gathered_ori_avg_reward_mean': '9.02941895', 'gathered_ori_avg_reward_std': '1.17931533'} +2026-06-27 04:59:46.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 04:59:46.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_0/index_1029_seed_33428_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 05:01:07.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:01:07.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:17.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_0/index_1029_seed_33429_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 05:02:39.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:02:39.194 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 05:02:40.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.572 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 252 train_steps 6080 {'step': '252/10000', 'learning_rate': '0.00001000', 'loss': '1.304e-08', 'grad_norm': '0.00431604', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.29971313', 'grpo_loop_time': '417.47070312', 'ref_model_time': '111.00276947', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.90679955', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.79492188', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.483e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00723418', 'kl_moving': '0.00723418', 'kl_stepwise': '-1.917e-08', 'gathered_avg_reward_mean': '9.79638672', 'gathered_avg_reward_std': '0.98117650', 'gathered_video_score2_mq_reward_mean': '3.31494141', 'gathered_video_score2_mq_reward_std': '0.40563184', 'gathered_video_score2_overall_reward_mean': '9.79492188', 'gathered_video_score2_overall_reward_std': '0.97631049', 'gathered_video_score2_ta_reward_mean': '3.32421875', 'gathered_video_score2_ta_reward_std': '0.44979581', 'gathered_video_score2_vq_reward_mean': '3.15722656', 'gathered_video_score2_vq_reward_std': '0.42612046', 'gathered_ori_avg_reward_mean': '9.79638672', 'gathered_ori_avg_reward_std': '0.98117650'} +2026-06-27 05:13:27.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 05:13:27.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:27.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:34.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_0/index_2212_seed_5570_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 05:14:50.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:50.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:15:57.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_0/index_2212_seed_5571_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:16:16.805 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 05:16:17.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.697 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 253 train_steps 6120 {'step': '253/10000', 'learning_rate': '0.00001000', 'loss': '6.985e-10', 'grad_norm': '0.00499594', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.90539551', 'grpo_loop_time': '418.98001099', 'ref_model_time': '112.08332825', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88216394', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23437500', 'token_credit_video_advantage_mean': '6.752e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.466e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.048e-08', 'kl_loss': '0.00799663', 'kl_moving': '0.00799663', 'kl_stepwise': '-1.928e-08', 'gathered_avg_reward_mean': '9.23132324', 'gathered_avg_reward_std': '0.98349452', 'gathered_video_score2_mq_reward_mean': '3.11962891', 'gathered_video_score2_mq_reward_std': '0.24983543', 'gathered_video_score2_overall_reward_mean': '9.23437500', 'gathered_video_score2_overall_reward_std': '0.98856682', 'gathered_video_score2_ta_reward_mean': '3.19201660', 'gathered_video_score2_ta_reward_std': '0.49208248', 'gathered_video_score2_vq_reward_mean': '2.91967773', 'gathered_video_score2_vq_reward_std': '0.46035671', 'gathered_ori_avg_reward_mean': '9.23132324', 'gathered_ori_avg_reward_std': '0.98349458'} +2026-06-27 05:27:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 05:27:07.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_0/index_9823_seed_31536_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 05:28:29.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:29.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:29:40.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_0/index_9823_seed_31537_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:30:00.697 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 05:30:01.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.235 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 254 train_steps 6160 {'step': '254/10000', 'learning_rate': '0.00001000', 'loss': '-6.054e-09', 'grad_norm': '0.01667659', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.72296143', 'grpo_loop_time': '418.11669922', 'ref_model_time': '111.81664276', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84404993', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.29296875', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.448e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.054e-09', 'kl_loss': '0.00791982', 'kl_moving': '0.00791982', 'kl_stepwise': '-2.024e-08', 'gathered_avg_reward_mean': '9.28601074', 'gathered_avg_reward_std': '1.20548689', 'gathered_video_score2_mq_reward_mean': '3.23437500', 'gathered_video_score2_mq_reward_std': '0.40032870', 'gathered_video_score2_overall_reward_mean': '9.29296875', 'gathered_video_score2_overall_reward_std': '1.20639896', 'gathered_video_score2_ta_reward_mean': '2.91943359', 'gathered_video_score2_ta_reward_std': '0.70647919', 'gathered_video_score2_vq_reward_mean': '3.13220215', 'gathered_video_score2_vq_reward_std': '0.53623742', 'gathered_ori_avg_reward_mean': '9.28601074', 'gathered_ori_avg_reward_std': '1.20548689'} +2026-06-27 05:40:50.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 05:40:50.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:56.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_0/index_2876_seed_14120_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 05:42:11.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:42:11.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:19.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_0/index_2876_seed_14121_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:43:43.873 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 05:43:44.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:34.904 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 255 train_steps 6200 {'step': '255/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00504042', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.75061035', 'grpo_loop_time': '419.13595581', 'ref_model_time': '111.57669067', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86072665', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.25585938', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.435e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00774507', 'kl_moving': '0.00774507', 'kl_stepwise': '-1.875e-08', 'gathered_avg_reward_mean': '9.25622559', 'gathered_avg_reward_std': '1.31181693', 'gathered_video_score2_mq_reward_mean': '3.26147461', 'gathered_video_score2_mq_reward_std': '0.41923764', 'gathered_video_score2_overall_reward_mean': '9.25585938', 'gathered_video_score2_overall_reward_std': '1.31319594', 'gathered_video_score2_ta_reward_mean': '2.93554688', 'gathered_video_score2_ta_reward_std': '0.64677936', 'gathered_video_score2_vq_reward_mean': '3.05920410', 'gathered_video_score2_vq_reward_std': '0.55765879', 'gathered_ori_avg_reward_mean': '9.25622559', 'gathered_ori_avg_reward_std': '1.31181693'} +2026-06-27 05:54:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 05:54:35.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_0/index_8512_seed_14125_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 05:55:54.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:54.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:07.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_0/index_8512_seed_14126_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:57:29.573 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 05:57:30.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.287 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 256 train_steps 6240 {'step': '256/10000', 'learning_rate': '0.00001000', 'loss': '1.490e-08', 'grad_norm': '0.00615502', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.52877808', 'grpo_loop_time': '418.14648438', 'ref_model_time': '111.31488800', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86730009', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.10156250', 'token_credit_video_advantage_mean': '-4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.468e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.490e-08', 'kl_loss': '0.00860841', 'kl_moving': '0.00860841', 'kl_stepwise': '-1.984e-08', 'gathered_avg_reward_mean': '9.10253906', 'gathered_avg_reward_std': '1.27853286', 'gathered_video_score2_mq_reward_mean': '3.12670898', 'gathered_video_score2_mq_reward_std': '0.37951100', 'gathered_video_score2_overall_reward_mean': '9.10156250', 'gathered_video_score2_overall_reward_std': '1.27851987', 'gathered_video_score2_ta_reward_mean': '3.09448242', 'gathered_video_score2_ta_reward_std': '0.67786485', 'gathered_video_score2_vq_reward_mean': '2.88134766', 'gathered_video_score2_vq_reward_std': '0.52098280', 'gathered_ori_avg_reward_mean': '9.10253906', 'gathered_ori_avg_reward_std': '1.27853286'} +2026-06-27 06:08:19.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 06:08:19.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:25.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_0/index_211_seed_8980_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 06:09:44.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:44.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:10:50.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_0/index_211_seed_8981_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:11:13.281 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 06:11:14.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.683 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 257 train_steps 6280 {'step': '257/10000', 'learning_rate': '0.00001000', 'loss': '-1.676e-08', 'grad_norm': '0.00657774', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.94494629', 'grpo_loop_time': '419.73272705', 'ref_model_time': '113.39105225', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85933137', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.47998047', 'token_credit_video_advantage_mean': '-3.027e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.407e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.049e-08', 'kl_loss': '0.00807729', 'kl_moving': '0.00807729', 'kl_stepwise': '-1.958e-08', 'gathered_avg_reward_mean': '9.47863770', 'gathered_avg_reward_std': '1.14365458', 'gathered_video_score2_mq_reward_mean': '3.30346680', 'gathered_video_score2_mq_reward_std': '0.40657669', 'gathered_video_score2_overall_reward_mean': '9.47998047', 'gathered_video_score2_overall_reward_std': '1.14626133', 'gathered_video_score2_ta_reward_mean': '3.10363770', 'gathered_video_score2_ta_reward_std': '0.59302074', 'gathered_video_score2_vq_reward_mean': '3.07153320', 'gathered_video_score2_vq_reward_std': '0.44056547', 'gathered_ori_avg_reward_mean': '9.47863770', 'gathered_ori_avg_reward_std': '1.14365458'} +2026-06-27 06:22:07.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 06:22:07.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:07.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:14.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_0/index_9845_seed_20996_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 06:23:28.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:28.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:39.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_0/index_9845_seed_20997_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:24:58.909 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 06:24:59.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.817 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 258 train_steps 6320 {'step': '258/10000', 'learning_rate': '0.00001000', 'loss': '-7.567e-09', 'grad_norm': '0.00767464', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.20019531', 'grpo_loop_time': '418.00537109', 'ref_model_time': '113.44853210', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84640038', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.31738281', 'token_credit_video_advantage_mean': '5.763e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.683e-09', 'kl_loss': '0.00874734', 'kl_moving': '0.00874734', 'kl_stepwise': '-2.058e-08', 'gathered_avg_reward_mean': '9.31701660', 'gathered_avg_reward_std': '0.78459221', 'gathered_video_score2_mq_reward_mean': '3.20458984', 'gathered_video_score2_mq_reward_std': '0.29906061', 'gathered_video_score2_overall_reward_mean': '9.31738281', 'gathered_video_score2_overall_reward_std': '0.78493696', 'gathered_video_score2_ta_reward_mean': '3.15698242', 'gathered_video_score2_ta_reward_std': '0.46257678', 'gathered_video_score2_vq_reward_mean': '2.95544434', 'gathered_video_score2_vq_reward_std': '0.43424359', 'gathered_ori_avg_reward_mean': '9.31701660', 'gathered_ori_avg_reward_std': '0.78459221'} +2026-06-27 06:35:49.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 06:35:49.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:56.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_0/index_6871_seed_32862_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 06:37:09.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:37:09.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:20.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_0/index_6871_seed_32863_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:38:41.759 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 06:38:42.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.712 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 259 train_steps 6360 {'step': '259/10000', 'learning_rate': '0.00001000', 'loss': '-6.054e-09', 'grad_norm': '0.00812841', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.84887695', 'grpo_loop_time': '418.50610352', 'ref_model_time': '111.33708191', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87998986', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.13769531', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.438e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.054e-09', 'kl_loss': '0.00922008', 'kl_moving': '0.00922008', 'kl_stepwise': '-1.932e-08', 'gathered_avg_reward_mean': '9.13879395', 'gathered_avg_reward_std': '0.89729750', 'gathered_video_score2_mq_reward_mean': '3.13159180', 'gathered_video_score2_mq_reward_std': '0.31711441', 'gathered_video_score2_overall_reward_mean': '9.13769531', 'gathered_video_score2_overall_reward_std': '0.89705598', 'gathered_video_score2_ta_reward_mean': '2.96984863', 'gathered_video_score2_ta_reward_std': '0.54193479', 'gathered_video_score2_vq_reward_mean': '3.03735352', 'gathered_video_score2_vq_reward_std': '0.34496430', 'gathered_ori_avg_reward_mean': '9.13879395', 'gathered_ori_avg_reward_std': '0.89729756'} +2026-06-27 06:49:30.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 06:49:30.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:30.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:36.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_0/index_118_seed_28482_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 06:50:52.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:52.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:02.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_0/index_118_seed_28483_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 06:52:20.916 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:52:21.087 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 06:52:22.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:12.587 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 260 train_steps 6400 {'step': '260/10000', 'learning_rate': '0.00001000', 'loss': '-1.523e-08', 'grad_norm': '0.00701173', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.72979736', 'grpo_loop_time': '418.75952148', 'ref_model_time': '111.81584167', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88063180', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.37011719', 'token_credit_video_advantage_mean': '-5.821e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.422e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.616e-08', 'kl_loss': '0.00940115', 'kl_moving': '0.00940115', 'kl_stepwise': '-1.982e-08', 'gathered_avg_reward_mean': '9.37316895', 'gathered_avg_reward_std': '1.18065453', 'gathered_video_score2_mq_reward_mean': '3.23144531', 'gathered_video_score2_mq_reward_std': '0.34621447', 'gathered_video_score2_overall_reward_mean': '9.37011719', 'gathered_video_score2_overall_reward_std': '1.17962563', 'gathered_video_score2_ta_reward_mean': '3.12670898', 'gathered_video_score2_ta_reward_std': '0.54844302', 'gathered_video_score2_vq_reward_mean': '3.01501465', 'gathered_video_score2_vq_reward_std': '0.47419989', 'gathered_ori_avg_reward_mean': '9.37316895', 'gathered_ori_avg_reward_std': '1.18065453'} +2026-06-27 07:03:47.826 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 07:03:49.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:57.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_0/index_4677_seed_31510_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 07:05:14.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:05:14.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:20.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_0/index_4677_seed_31511_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:06:39.760 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 07:06:40.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.099 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 261 train_steps 6440 {'step': '261/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.01126826', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.78433228', 'grpo_loop_time': '418.14816284', 'ref_model_time': '111.71134949', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86020732', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.16113281', 'token_credit_video_advantage_mean': '-2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.436e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00877523', 'kl_moving': '0.00877523', 'kl_stepwise': '-1.850e-08', 'gathered_avg_reward_mean': '9.16162109', 'gathered_avg_reward_std': '1.01445329', 'gathered_video_score2_mq_reward_mean': '3.17285156', 'gathered_video_score2_mq_reward_std': '0.29015413', 'gathered_video_score2_overall_reward_mean': '9.16113281', 'gathered_video_score2_overall_reward_std': '1.01382375', 'gathered_video_score2_ta_reward_mean': '3.10839844', 'gathered_video_score2_ta_reward_std': '0.50148457', 'gathered_video_score2_vq_reward_mean': '2.88037109', 'gathered_video_score2_vq_reward_std': '0.44914418', 'gathered_ori_avg_reward_mean': '9.16162109', 'gathered_ori_avg_reward_std': '1.01445329'} +2026-06-27 07:17:29.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 07:17:29.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:35.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_0/index_8624_seed_17155_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 07:18:47.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:47.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:00.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_0/index_8624_seed_17156_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:20:21.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 07:20:23.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.061 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 262 train_steps 6480 {'step': '262/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00568398', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.13201904', 'grpo_loop_time': '420.75360107', 'ref_model_time': '111.70786285', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85358745', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.16015625', 'token_credit_video_advantage_mean': '3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.472e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00811997', 'kl_moving': '0.00811997', 'kl_stepwise': '-1.958e-08', 'gathered_avg_reward_mean': '9.16003418', 'gathered_avg_reward_std': '0.89894265', 'gathered_video_score2_mq_reward_mean': '3.15283203', 'gathered_video_score2_mq_reward_std': '0.31606209', 'gathered_video_score2_overall_reward_mean': '9.16015625', 'gathered_video_score2_overall_reward_std': '0.89621836', 'gathered_video_score2_ta_reward_mean': '3.02978516', 'gathered_video_score2_ta_reward_std': '0.63569623', 'gathered_video_score2_vq_reward_mean': '2.97741699', 'gathered_video_score2_vq_reward_std': '0.35850650', 'gathered_ori_avg_reward_mean': '9.16003418', 'gathered_ori_avg_reward_std': '0.89894265'} +2026-06-27 07:31:16.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 07:31:16.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:22.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_0/index_3190_seed_19671_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 07:32:35.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:35.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:33:49.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_0/index_3190_seed_19672_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:34:11.330 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 07:34:12.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.293 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 263 train_steps 6520 {'step': '263/10000', 'learning_rate': '0.00001000', 'loss': '-2.980e-09', 'grad_norm': '0.00658566', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.37322998', 'grpo_loop_time': '420.62213135', 'ref_model_time': '113.51618195', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85595238', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.12597656', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.364e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.980e-09', 'kl_loss': '0.00850507', 'kl_moving': '0.00850507', 'kl_stepwise': '-1.859e-08', 'gathered_avg_reward_mean': '9.12609863', 'gathered_avg_reward_std': '0.85592222', 'gathered_video_score2_mq_reward_mean': '3.12207031', 'gathered_video_score2_mq_reward_std': '0.25908002', 'gathered_video_score2_overall_reward_mean': '9.12597656', 'gathered_video_score2_overall_reward_std': '0.85805953', 'gathered_video_score2_ta_reward_mean': '3.10583496', 'gathered_video_score2_ta_reward_std': '0.51921898', 'gathered_video_score2_vq_reward_mean': '2.89819336', 'gathered_video_score2_vq_reward_std': '0.47321072', 'gathered_ori_avg_reward_mean': '9.12609863', 'gathered_ori_avg_reward_std': '0.85592222'} +2026-06-27 07:45:07.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 07:45:07.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:13.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_0/index_1399_seed_15704_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 07:46:26.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:26.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:47:38.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_0/index_1399_seed_15705_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 07:48:01.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:48:01.406 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 07:48:02.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.098 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 264 train_steps 6560 {'step': '264/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00547646', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.76095581', 'grpo_loop_time': '418.44869995', 'ref_model_time': '112.88567352', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87024766', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.08398438', 'token_credit_video_advantage_mean': '-3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.372e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.382e-09', 'kl_loss': '0.00811965', 'kl_moving': '0.00811965', 'kl_stepwise': '-1.857e-08', 'gathered_avg_reward_mean': '9.08422852', 'gathered_avg_reward_std': '1.14789963', 'gathered_video_score2_mq_reward_mean': '3.16967773', 'gathered_video_score2_mq_reward_std': '0.46070462', 'gathered_video_score2_overall_reward_mean': '9.08398438', 'gathered_video_score2_overall_reward_std': '1.15327525', 'gathered_video_score2_ta_reward_mean': '2.92431641', 'gathered_video_score2_ta_reward_std': '0.58312291', 'gathered_video_score2_vq_reward_mean': '2.99023438', 'gathered_video_score2_vq_reward_std': '0.53540212', 'gathered_ori_avg_reward_mean': '9.08422852', 'gathered_ori_avg_reward_std': '1.14789963'} +2026-06-27 07:58:52.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 07:58:52.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_0/index_2623_seed_17235_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 08:00:18.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:00:18.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:23.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_0/index_2623_seed_17236_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 08:01:45.186 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:01:45.281 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 08:01:46.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.829 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 265 train_steps 6600 {'step': '265/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00442980', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.76150513', 'grpo_loop_time': '419.22872925', 'ref_model_time': '111.75329590', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87693453', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.95019531', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.433e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00788614', 'kl_moving': '0.00788614', 'kl_stepwise': '-1.933e-08', 'gathered_avg_reward_mean': '8.94909668', 'gathered_avg_reward_std': '1.37492573', 'gathered_video_score2_mq_reward_mean': '3.06274414', 'gathered_video_score2_mq_reward_std': '0.53468198', 'gathered_video_score2_overall_reward_mean': '8.95019531', 'gathered_video_score2_overall_reward_std': '1.37539637', 'gathered_video_score2_ta_reward_mean': '2.90771484', 'gathered_video_score2_ta_reward_std': '0.63346905', 'gathered_video_score2_vq_reward_mean': '2.97863770', 'gathered_video_score2_vq_reward_std': '0.51196331', 'gathered_ori_avg_reward_mean': '8.94909668', 'gathered_ori_avg_reward_std': '1.37492573'} +2026-06-27 08:12:37.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 08:12:37.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_0/index_5670_seed_16870_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 08:13:57.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:57.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:10.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_0/index_5670_seed_16871_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:15:30.814 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 08:15:31.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.483 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 266 train_steps 6640 {'step': '266/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00562587', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.84362793', 'grpo_loop_time': '418.83157349', 'ref_model_time': '111.26440430', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84633380', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.67480469', 'token_credit_video_advantage_mean': '-2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00843830', 'kl_moving': '0.00843830', 'kl_stepwise': '-1.912e-08', 'gathered_avg_reward_mean': '8.67492676', 'gathered_avg_reward_std': '0.85480142', 'gathered_video_score2_mq_reward_mean': '3.06689453', 'gathered_video_score2_mq_reward_std': '0.25948599', 'gathered_video_score2_overall_reward_mean': '8.67480469', 'gathered_video_score2_overall_reward_std': '0.85291821', 'gathered_video_score2_ta_reward_mean': '2.82482910', 'gathered_video_score2_ta_reward_std': '0.58438891', 'gathered_video_score2_vq_reward_mean': '2.78320312', 'gathered_video_score2_vq_reward_std': '0.45927310', 'gathered_ori_avg_reward_mean': '8.67492676', 'gathered_ori_avg_reward_std': '0.85480142'} +2026-06-27 08:26:20.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 08:26:20.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:26.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_0/index_9254_seed_15485_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 08:27:39.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:39.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:28:51.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_0/index_9254_seed_15486_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:29:10.191 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 08:29:11.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.745 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 267 train_steps 6680 {'step': '267/10000', 'learning_rate': '0.00001000', 'loss': '5.122e-09', 'grad_norm': '0.00821296', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.16378784', 'grpo_loop_time': '421.29678345', 'ref_model_time': '111.96359253', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86309969', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24267578', 'token_credit_video_advantage_mean': '-8.324e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.433e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.122e-09', 'kl_loss': '0.00895909', 'kl_moving': '0.00895909', 'kl_stepwise': '-1.977e-08', 'gathered_avg_reward_mean': '9.24096680', 'gathered_avg_reward_std': '0.99250865', 'gathered_video_score2_mq_reward_mean': '3.11816406', 'gathered_video_score2_mq_reward_std': '0.32981655', 'gathered_video_score2_overall_reward_mean': '9.24267578', 'gathered_video_score2_overall_reward_std': '0.99715388', 'gathered_video_score2_ta_reward_mean': '3.20678711', 'gathered_video_score2_ta_reward_std': '0.49876052', 'gathered_video_score2_vq_reward_mean': '2.91601562', 'gathered_video_score2_vq_reward_std': '0.43537825', 'gathered_ori_avg_reward_mean': '9.24096680', 'gathered_ori_avg_reward_std': '0.99250871'} +2026-06-27 08:40:02.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 08:40:02.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:02.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_0/index_6713_seed_23219_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 08:41:23.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:23.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:35.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_0/index_6713_seed_23220_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 08:42:56.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:42:57.099 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 08:42:57.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.719 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 268 train_steps 6720 {'step': '268/10000', 'learning_rate': '0.00001000', 'loss': '1.071e-08', 'grad_norm': '0.00510306', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.84100342', 'grpo_loop_time': '418.15655518', 'ref_model_time': '112.98004913', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87168431', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.40380859', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.446e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00903295', 'kl_moving': '0.00903295', 'kl_stepwise': '-2.039e-08', 'gathered_avg_reward_mean': '9.39770508', 'gathered_avg_reward_std': '0.99030918', 'gathered_video_score2_mq_reward_mean': '3.16345215', 'gathered_video_score2_mq_reward_std': '0.36671269', 'gathered_video_score2_overall_reward_mean': '9.40380859', 'gathered_video_score2_overall_reward_std': '0.99093032', 'gathered_video_score2_ta_reward_mean': '3.20056152', 'gathered_video_score2_ta_reward_std': '0.56663364', 'gathered_video_score2_vq_reward_mean': '3.03369141', 'gathered_video_score2_vq_reward_std': '0.29438743', 'gathered_ori_avg_reward_mean': '9.39770508', 'gathered_ori_avg_reward_std': '0.99030924'} +2026-06-27 08:53:47.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 08:53:47.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:47.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:54.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_0/index_6348_seed_27062_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 08:55:10.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:55:10.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:23.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_0/index_6348_seed_27063_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 08:56:49.418 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:56:49.584 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 08:56:50.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.538 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 269 train_steps 6760 {'step': '269/10000', 'learning_rate': '0.00001000', 'loss': '1.676e-08', 'grad_norm': '0.00647170', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '295.03164673', 'grpo_loop_time': '418.85745239', 'ref_model_time': '111.60657501', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85394299', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.45458984', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.437e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.676e-08', 'kl_loss': '0.00923856', 'kl_moving': '0.00923856', 'kl_stepwise': '-1.911e-08', 'gathered_avg_reward_mean': '9.45544434', 'gathered_avg_reward_std': '0.98977166', 'gathered_video_score2_mq_reward_mean': '3.15490723', 'gathered_video_score2_mq_reward_std': '0.38680148', 'gathered_video_score2_overall_reward_mean': '9.45458984', 'gathered_video_score2_overall_reward_std': '0.99315482', 'gathered_video_score2_ta_reward_mean': '3.30932617', 'gathered_video_score2_ta_reward_std': '0.51476741', 'gathered_video_score2_vq_reward_mean': '2.99121094', 'gathered_video_score2_vq_reward_std': '0.39318377', 'gathered_ori_avg_reward_mean': '9.45544434', 'gathered_ori_avg_reward_std': '0.98977166'} +2026-06-27 09:07:41.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 09:07:41.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:47.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_0/index_2266_seed_11959_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 09:09:02.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:09:02.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:14.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_0/index_2266_seed_11960_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:10:36.648 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 09:10:37.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:21:25.697 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 270 train_steps 6800 {'step': '270/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00510165', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.93811035', 'grpo_loop_time': '418.41006470', 'ref_model_time': '111.24186707', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84422463', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.29736328', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.406e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00933365', 'kl_moving': '0.00933366', 'kl_stepwise': '-1.871e-08', 'gathered_avg_reward_mean': '9.29589844', 'gathered_avg_reward_std': '1.02741468', 'gathered_video_score2_mq_reward_mean': '3.20288086', 'gathered_video_score2_mq_reward_std': '0.32853737', 'gathered_video_score2_overall_reward_mean': '9.29736328', 'gathered_video_score2_overall_reward_std': '1.02669740', 'gathered_video_score2_ta_reward_mean': '3.00598145', 'gathered_video_score2_ta_reward_std': '0.60123861', 'gathered_video_score2_vq_reward_mean': '3.08703613', 'gathered_video_score2_vq_reward_std': '0.41148609', 'gathered_ori_avg_reward_mean': '9.29589844', 'gathered_ori_avg_reward_std': '1.02741480'} +2026-06-27 09:22:00.983 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 09:22:02.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:10.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_0/index_4261_seed_6135_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 09:23:24.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:24.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:34.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_0/index_4261_seed_6136_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:24:54.842 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 09:24:55.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.131 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 271 train_steps 6840 {'step': '271/10000', 'learning_rate': '0.00001000', 'loss': '-3.260e-09', 'grad_norm': '0.00521315', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.30859375', 'grpo_loop_time': '418.27679443', 'ref_model_time': '110.67932129', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86027294', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.30078125', 'token_credit_video_advantage_mean': '4.889e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.260e-09', 'kl_loss': '0.00810925', 'kl_moving': '0.00810925', 'kl_stepwise': '-1.912e-08', 'gathered_avg_reward_mean': '9.30590820', 'gathered_avg_reward_std': '0.96234798', 'gathered_video_score2_mq_reward_mean': '3.24707031', 'gathered_video_score2_mq_reward_std': '0.37344930', 'gathered_video_score2_overall_reward_mean': '9.30078125', 'gathered_video_score2_overall_reward_std': '0.96197701', 'gathered_video_score2_ta_reward_mean': '3.00415039', 'gathered_video_score2_ta_reward_std': '0.62471914', 'gathered_video_score2_vq_reward_mean': '3.05468750', 'gathered_video_score2_vq_reward_std': '0.45192796', 'gathered_ori_avg_reward_mean': '9.30590820', 'gathered_ori_avg_reward_std': '0.96234792'} +2026-06-27 09:35:43.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 09:35:43.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:49.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_0/index_8454_seed_19743_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 09:37:04.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:37:04.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:13.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_0/index_8454_seed_19744_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:38:34.471 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 09:38:35.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.189 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.831 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 272 train_steps 6880 {'step': '272/10000', 'learning_rate': '0.00001000', 'loss': '3.260e-09', 'grad_norm': '0.00470552', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.78125000', 'grpo_loop_time': '419.98437500', 'ref_model_time': '111.88682556', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85870159', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20068359', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.450e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.260e-09', 'kl_loss': '0.00806868', 'kl_moving': '0.00806869', 'kl_stepwise': '-1.950e-08', 'gathered_avg_reward_mean': '9.20056152', 'gathered_avg_reward_std': '1.03002298', 'gathered_video_score2_mq_reward_mean': '3.19848633', 'gathered_video_score2_mq_reward_std': '0.36049503', 'gathered_video_score2_overall_reward_mean': '9.20068359', 'gathered_video_score2_overall_reward_std': '1.02918839', 'gathered_video_score2_ta_reward_mean': '3.02795410', 'gathered_video_score2_ta_reward_std': '0.61889076', 'gathered_video_score2_vq_reward_mean': '2.97412109', 'gathered_video_score2_vq_reward_std': '0.33750337', 'gathered_ori_avg_reward_mean': '9.20056152', 'gathered_ori_avg_reward_std': '1.03002310'} +2026-06-27 09:49:25.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 09:49:25.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_0/index_8254_seed_31702_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 09:50:44.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:44.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:51:58.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_0/index_8254_seed_31703_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:52:16.995 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 09:52:18.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.931 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 273 train_steps 6920 {'step': '273/10000', 'learning_rate': '0.00001000', 'loss': '-1.024e-08', 'grad_norm': '0.00602677', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.92709351', 'grpo_loop_time': '419.01220703', 'ref_model_time': '111.94496918', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84582579', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.34570312', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.442e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.024e-08', 'kl_loss': '0.00805602', 'kl_moving': '0.00805602', 'kl_stepwise': '-1.885e-08', 'gathered_avg_reward_mean': '9.34716797', 'gathered_avg_reward_std': '1.12892807', 'gathered_video_score2_mq_reward_mean': '3.22387695', 'gathered_video_score2_mq_reward_std': '0.39612395', 'gathered_video_score2_overall_reward_mean': '9.34570312', 'gathered_video_score2_overall_reward_std': '1.12568986', 'gathered_video_score2_ta_reward_mean': '3.07275391', 'gathered_video_score2_ta_reward_std': '0.58558077', 'gathered_video_score2_vq_reward_mean': '3.05053711', 'gathered_video_score2_vq_reward_std': '0.47167659', 'gathered_ori_avg_reward_mean': '9.34716797', 'gathered_ori_avg_reward_std': '1.12892807'} +2026-06-27 10:03:07.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 10:03:08.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_0/index_4468_seed_16539_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 10:04:27.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:27.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:39.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_0/index_4468_seed_16540_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:06:00.029 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 10:06:00.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.908 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 274 train_steps 6960 {'step': '274/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00425134', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.71963501', 'grpo_loop_time': '419.78320312', 'ref_model_time': '113.05483246', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83778763', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24951172', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00777572', 'kl_moving': '0.00777572', 'kl_stepwise': '-1.936e-08', 'gathered_avg_reward_mean': '9.25036621', 'gathered_avg_reward_std': '1.06054461', 'gathered_video_score2_mq_reward_mean': '3.18945312', 'gathered_video_score2_mq_reward_std': '0.34597889', 'gathered_video_score2_overall_reward_mean': '9.24951172', 'gathered_video_score2_overall_reward_std': '1.06059527', 'gathered_video_score2_ta_reward_mean': '3.01782227', 'gathered_video_score2_ta_reward_std': '0.68843776', 'gathered_video_score2_vq_reward_mean': '3.04309082', 'gathered_video_score2_vq_reward_std': '0.36831036', 'gathered_ori_avg_reward_mean': '9.25036621', 'gathered_ori_avg_reward_std': '1.06054461'} +2026-06-27 10:16:52.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 10:16:53.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_0/index_2797_seed_13962_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 10:18:13.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:18:13.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:24.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_0/index_2797_seed_13963_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 10:19:45.587 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:19:45.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 10:19:46.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:30:39.337 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 275 train_steps 7000 {'step': '275/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-10', 'grad_norm': '0.00905960', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.13476562', 'grpo_loop_time': '420.06277466', 'ref_model_time': '111.56816864', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.81202680', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.19726562', 'token_credit_video_advantage_mean': '7.101e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00786690', 'kl_moving': '0.00786691', 'kl_stepwise': '-1.976e-08', 'gathered_avg_reward_mean': '9.19921875', 'gathered_avg_reward_std': '1.30472410', 'gathered_video_score2_mq_reward_mean': '3.18115234', 'gathered_video_score2_mq_reward_std': '0.43421549', 'gathered_video_score2_overall_reward_mean': '9.19726562', 'gathered_video_score2_overall_reward_std': '1.30440533', 'gathered_video_score2_ta_reward_mean': '2.87927246', 'gathered_video_score2_ta_reward_std': '0.71339339', 'gathered_video_score2_vq_reward_mean': '3.13879395', 'gathered_video_score2_vq_reward_std': '0.48117632', 'gathered_ori_avg_reward_mean': '9.19921875', 'gathered_ori_avg_reward_std': '1.30472410'} +2026-06-27 10:31:46.501 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 275 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-275 +2026-06-27 10:31:46.638 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 275: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-275 +2026-06-27 10:31:49.732 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-250 +2026-06-27 10:31:49.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 10:31:49.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:49.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_0/index_5551_seed_26851_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 10:33:11.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:33:11.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:21.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_0/index_5551_seed_26852_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:34:42.934 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 10:34:43.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.682 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 276 train_steps 7040 {'step': '276/10000', 'learning_rate': '0.00001000', 'loss': '-8.382e-09', 'grad_norm': '0.00502242', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.38287354', 'grpo_loop_time': '417.64547729', 'ref_model_time': '111.54138184', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85898662', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23193359', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.414e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.00864897', 'kl_moving': '0.00864897', 'kl_stepwise': '-1.988e-08', 'gathered_avg_reward_mean': '9.23461914', 'gathered_avg_reward_std': '1.14262652', 'gathered_video_score2_mq_reward_mean': '3.14221191', 'gathered_video_score2_mq_reward_std': '0.48922014', 'gathered_video_score2_overall_reward_mean': '9.23193359', 'gathered_video_score2_overall_reward_std': '1.14234674', 'gathered_video_score2_ta_reward_mean': '2.95593262', 'gathered_video_score2_ta_reward_std': '0.66397542', 'gathered_video_score2_vq_reward_mean': '3.13647461', 'gathered_video_score2_vq_reward_std': '0.40262091', 'gathered_ori_avg_reward_mean': '9.23461914', 'gathered_ori_avg_reward_std': '1.14262652'} +2026-06-27 10:45:31.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 10:45:31.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:38.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_0/index_786_seed_8818_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 10:46:54.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:54.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:03.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_0/index_786_seed_8819_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:48:26.520 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 10:48:27.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.410 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 277 train_steps 7080 {'step': '277/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-10', 'grad_norm': '0.00566599', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.89746094', 'grpo_loop_time': '417.22482300', 'ref_model_time': '111.48675537', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86506110', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.28515625', 'token_credit_video_advantage_mean': '-3.958e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.435e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00913278', 'kl_moving': '0.00913278', 'kl_stepwise': '-1.830e-08', 'gathered_avg_reward_mean': '9.29016113', 'gathered_avg_reward_std': '1.17615902', 'gathered_video_score2_mq_reward_mean': '3.15051270', 'gathered_video_score2_mq_reward_std': '0.41255632', 'gathered_video_score2_overall_reward_mean': '9.28515625', 'gathered_video_score2_overall_reward_std': '1.17770040', 'gathered_video_score2_ta_reward_mean': '3.13208008', 'gathered_video_score2_ta_reward_std': '0.69213927', 'gathered_video_score2_vq_reward_mean': '3.00756836', 'gathered_video_score2_vq_reward_std': '0.45114222', 'gathered_ori_avg_reward_mean': '9.29016113', 'gathered_ori_avg_reward_std': '1.17615902'} +2026-06-27 10:59:16.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 10:59:16.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_0/index_2420_seed_29966_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 11:00:35.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:35.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:01:49.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_0/index_2420_seed_29967_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:02:12.225 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 11:02:13.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.836 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 278 train_steps 7120 {'step': '278/10000', 'learning_rate': '0.00001000', 'loss': '8.382e-09', 'grad_norm': '0.00637259', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.40097046', 'grpo_loop_time': '419.84011841', 'ref_model_time': '112.99537659', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88376623', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.74902344', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.00971166', 'kl_moving': '0.00971166', 'kl_stepwise': '-1.911e-08', 'gathered_avg_reward_mean': '9.74621582', 'gathered_avg_reward_std': '1.18939137', 'gathered_video_score2_mq_reward_mean': '3.30688477', 'gathered_video_score2_mq_reward_std': '0.41077304', 'gathered_video_score2_overall_reward_mean': '9.74902344', 'gathered_video_score2_overall_reward_std': '1.19276905', 'gathered_video_score2_ta_reward_mean': '3.29785156', 'gathered_video_score2_ta_reward_std': '0.54370505', 'gathered_video_score2_vq_reward_mean': '3.14147949', 'gathered_video_score2_vq_reward_std': '0.43240467', 'gathered_ori_avg_reward_mean': '9.74621582', 'gathered_ori_avg_reward_std': '1.18939137'} +2026-06-27 11:13:06.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 11:13:07.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:13.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_0/index_8420_seed_11922_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 11:14:27.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:27.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:36.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_0/index_8420_seed_11923_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:15:55.198 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 11:15:55.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.016 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 279 train_steps 7160 {'step': '279/10000', 'learning_rate': '0.00001000', 'loss': '-8.382e-09', 'grad_norm': '0.00461107', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.16015625', 'grpo_loop_time': '418.70422363', 'ref_model_time': '113.25871277', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86517334', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.41552734', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.476e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.00914952', 'kl_moving': '0.00914953', 'kl_stepwise': '-1.958e-08', 'gathered_avg_reward_mean': '9.41638184', 'gathered_avg_reward_std': '1.03900349', 'gathered_video_score2_mq_reward_mean': '3.19506836', 'gathered_video_score2_mq_reward_std': '0.32689026', 'gathered_video_score2_overall_reward_mean': '9.41552734', 'gathered_video_score2_overall_reward_std': '1.03523254', 'gathered_video_score2_ta_reward_mean': '3.12011719', 'gathered_video_score2_ta_reward_std': '0.55109370', 'gathered_video_score2_vq_reward_mean': '3.10119629', 'gathered_video_score2_vq_reward_std': '0.42623666', 'gathered_ori_avg_reward_mean': '9.41638184', 'gathered_ori_avg_reward_std': '1.03900349'} +2026-06-27 11:26:47.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 11:26:47.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:53.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_0/index_7617_seed_32476_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 11:28:10.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:28:10.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:21.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_0/index_7617_seed_32477_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 11:29:40.798 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:29:40.982 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 11:29:42.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:40:29.216 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 280 train_steps 7200 {'step': '280/10000', 'learning_rate': '0.00001000', 'loss': '-3.469e-09', 'grad_norm': '0.00468202', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.14038086', 'grpo_loop_time': '417.01806641', 'ref_model_time': '111.40110779', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87423664', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.90185547', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.408e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.378e-09', 'kl_loss': '0.00962103', 'kl_moving': '0.00962103', 'kl_stepwise': '-1.999e-08', 'gathered_avg_reward_mean': '8.90039062', 'gathered_avg_reward_std': '0.90922010', 'gathered_video_score2_mq_reward_mean': '3.06958008', 'gathered_video_score2_mq_reward_std': '0.32260197', 'gathered_video_score2_overall_reward_mean': '8.90185547', 'gathered_video_score2_overall_reward_std': '0.90752649', 'gathered_video_score2_ta_reward_mean': '2.92932129', 'gathered_video_score2_ta_reward_std': '0.53730363', 'gathered_video_score2_vq_reward_mean': '2.90148926', 'gathered_video_score2_vq_reward_std': '0.46348074', 'gathered_ori_avg_reward_mean': '8.90039062', 'gathered_ori_avg_reward_std': '0.90922010'} +2026-06-27 11:41:04.344 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 11:41:06.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:13.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_0/index_9409_seed_10560_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 11:42:29.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:30.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:38.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_0/index_9409_seed_10561_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:43:58.293 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 11:43:59.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.687 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 281 train_steps 7240 {'step': '281/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00591921', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.07516479', 'grpo_loop_time': '416.76687622', 'ref_model_time': '113.10065460', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85686457', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.56689453', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.461e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00872860', 'kl_moving': '0.00872861', 'kl_stepwise': '-2.014e-08', 'gathered_avg_reward_mean': '9.56799316', 'gathered_avg_reward_std': '1.01511288', 'gathered_video_score2_mq_reward_mean': '3.30346680', 'gathered_video_score2_mq_reward_std': '0.34520957', 'gathered_video_score2_overall_reward_mean': '9.56689453', 'gathered_video_score2_overall_reward_std': '1.01923740', 'gathered_video_score2_ta_reward_mean': '3.28576660', 'gathered_video_score2_ta_reward_std': '0.60009718', 'gathered_video_score2_vq_reward_mean': '2.97875977', 'gathered_video_score2_vq_reward_std': '0.51759744', 'gathered_ori_avg_reward_mean': '9.56799316', 'gathered_ori_avg_reward_std': '1.01511288'} +2026-06-27 11:54:47.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 11:54:48.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_0/index_2505_seed_5406_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 11:56:14.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:56:14.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:20.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_0/index_2505_seed_5407_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:57:44.633 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 11:57:45.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.802 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 282 train_steps 7280 {'step': '282/10000', 'learning_rate': '0.00001000', 'loss': '1.397e-08', 'grad_norm': '0.00502989', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.43630981', 'grpo_loop_time': '417.14068604', 'ref_model_time': '110.98550415', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87870884', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.17285156', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.437e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.397e-08', 'kl_loss': '0.00821521', 'kl_moving': '0.00821521', 'kl_stepwise': '-1.905e-08', 'gathered_avg_reward_mean': '9.17419434', 'gathered_avg_reward_std': '1.03721750', 'gathered_video_score2_mq_reward_mean': '3.19531250', 'gathered_video_score2_mq_reward_std': '0.36913547', 'gathered_video_score2_overall_reward_mean': '9.17285156', 'gathered_video_score2_overall_reward_std': '1.04458368', 'gathered_video_score2_ta_reward_mean': '2.97241211', 'gathered_video_score2_ta_reward_std': '0.49725154', 'gathered_video_score2_vq_reward_mean': '3.00646973', 'gathered_video_score2_vq_reward_std': '0.52390110', 'gathered_ori_avg_reward_mean': '9.17419434', 'gathered_ori_avg_reward_std': '1.03721762'} +2026-06-27 12:08:33.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 12:08:34.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:40.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_0/index_3757_seed_13202_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 12:10:00.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:10:00.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:06.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_0/index_3757_seed_13203_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:11:26.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 12:11:27.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.338 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 283 train_steps 7320 {'step': '283/10000', 'learning_rate': '0.00001000', 'loss': '1.304e-09', 'grad_norm': '0.00423273', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.24520874', 'grpo_loop_time': '418.80325317', 'ref_model_time': '111.58992767', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86670041', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.19042969', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.346e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.304e-09', 'kl_loss': '0.00809988', 'kl_moving': '0.00809989', 'kl_stepwise': '-1.905e-08', 'gathered_avg_reward_mean': '9.18920898', 'gathered_avg_reward_std': '1.09689879', 'gathered_video_score2_mq_reward_mean': '3.20690918', 'gathered_video_score2_mq_reward_std': '0.39736614', 'gathered_video_score2_overall_reward_mean': '9.19042969', 'gathered_video_score2_overall_reward_std': '1.09356475', 'gathered_video_score2_ta_reward_mean': '2.91540527', 'gathered_video_score2_ta_reward_std': '0.64217216', 'gathered_video_score2_vq_reward_mean': '3.06689453', 'gathered_video_score2_vq_reward_std': '0.44449338', 'gathered_ori_avg_reward_mean': '9.18920898', 'gathered_ori_avg_reward_std': '1.09689879'} +2026-06-27 12:22:19.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 12:22:19.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:25.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_0/index_9857_seed_1086_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 12:23:39.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:39.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:24:50.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_0/index_9857_seed_1087_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 12:25:09.431 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:25:09.866 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 12:25:10.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.059 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 284 train_steps 7360 {'step': '284/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.00516566', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.64837646', 'grpo_loop_time': '419.69772339', 'ref_model_time': '111.56491852', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86480582', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.36376953', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.476e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.794e-09', 'kl_loss': '0.00834960', 'kl_moving': '0.00834960', 'kl_stepwise': '-1.998e-08', 'gathered_avg_reward_mean': '9.36437988', 'gathered_avg_reward_std': '0.91493571', 'gathered_video_score2_mq_reward_mean': '3.19262695', 'gathered_video_score2_mq_reward_std': '0.30933523', 'gathered_video_score2_overall_reward_mean': '9.36376953', 'gathered_video_score2_overall_reward_std': '0.91693538', 'gathered_video_score2_ta_reward_mean': '2.97302246', 'gathered_video_score2_ta_reward_std': '0.65899509', 'gathered_video_score2_vq_reward_mean': '3.19873047', 'gathered_video_score2_vq_reward_std': '0.32153052', 'gathered_ori_avg_reward_mean': '9.36437988', 'gathered_ori_avg_reward_std': '0.91493577'} +2026-06-27 12:36:01.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 12:36:01.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:07.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_0/index_9587_seed_30337_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 12:37:25.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:25.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:33.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_0/index_9587_seed_30338_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:38:52.625 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 12:38:53.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.070 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 285 train_steps 7400 {'step': '285/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00623306', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.35083008', 'grpo_loop_time': '419.38183594', 'ref_model_time': '111.28494263', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86460888', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.97265625', 'token_credit_video_advantage_mean': '9.313e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00955379', 'kl_moving': '0.00955379', 'kl_stepwise': '-2.005e-08', 'gathered_avg_reward_mean': '8.97338867', 'gathered_avg_reward_std': '0.96555483', 'gathered_video_score2_mq_reward_mean': '3.07983398', 'gathered_video_score2_mq_reward_std': '0.21752793', 'gathered_video_score2_overall_reward_mean': '8.97265625', 'gathered_video_score2_overall_reward_std': '0.96688169', 'gathered_video_score2_ta_reward_mean': '3.11303711', 'gathered_video_score2_ta_reward_std': '0.46122494', 'gathered_video_score2_vq_reward_mean': '2.78051758', 'gathered_video_score2_vq_reward_std': '0.48927772', 'gathered_ori_avg_reward_mean': '8.97338867', 'gathered_ori_avg_reward_std': '0.96555489'} +2026-06-27 12:49:44.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 12:49:44.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:50.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_0/index_8410_seed_22470_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 12:51:04.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:51:04.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:15.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_0/index_8410_seed_22471_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:52:34.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 12:52:35.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.239 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 286 train_steps 7440 {'step': '286/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00541869', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.76986694', 'grpo_loop_time': '416.84552002', 'ref_model_time': '111.16182709', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85442781', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.54150391', 'token_credit_video_advantage_mean': '4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.406e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00889356', 'kl_moving': '0.00889356', 'kl_stepwise': '-1.852e-08', 'gathered_avg_reward_mean': '9.54443359', 'gathered_avg_reward_std': '1.24921608', 'gathered_video_score2_mq_reward_mean': '3.23974609', 'gathered_video_score2_mq_reward_std': '0.43723154', 'gathered_video_score2_overall_reward_mean': '9.54150391', 'gathered_video_score2_overall_reward_std': '1.24970770', 'gathered_video_score2_ta_reward_mean': '3.14562988', 'gathered_video_score2_ta_reward_std': '0.60834181', 'gathered_video_score2_vq_reward_mean': '3.15905762', 'gathered_video_score2_vq_reward_std': '0.50815487', 'gathered_ori_avg_reward_mean': '9.54443359', 'gathered_ori_avg_reward_std': '1.24921608'} +2026-06-27 13:03:21.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 13:03:21.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:27.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_0/index_829_seed_21057_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 13:04:41.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:41.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:05:51.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_0/index_829_seed_21058_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:06:12.746 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 13:06:13.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.620 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 287 train_steps 7480 {'step': '287/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00647319', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.83251953', 'grpo_loop_time': '420.01452637', 'ref_model_time': '112.74649048', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87062812', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.28271484', 'token_credit_video_advantage_mean': '2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.448e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.00939986', 'kl_moving': '0.00939986', 'kl_stepwise': '-2.016e-08', 'gathered_avg_reward_mean': '9.28100586', 'gathered_avg_reward_std': '1.00864577', 'gathered_video_score2_mq_reward_mean': '3.18310547', 'gathered_video_score2_mq_reward_std': '0.32189810', 'gathered_video_score2_overall_reward_mean': '9.28271484', 'gathered_video_score2_overall_reward_std': '1.01227558', 'gathered_video_score2_ta_reward_mean': '3.18041992', 'gathered_video_score2_ta_reward_std': '0.54006064', 'gathered_video_score2_vq_reward_mean': '2.91748047', 'gathered_video_score2_vq_reward_std': '0.48169360', 'gathered_ori_avg_reward_mean': '9.28100586', 'gathered_ori_avg_reward_std': '1.00864577'} +2026-06-27 13:17:07.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 13:17:07.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:07.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_0/index_8441_seed_16534_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 13:18:28.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:28.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:37.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_0/index_8441_seed_16535_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:19:57.758 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 13:19:58.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.003 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 288 train_steps 7520 {'step': '288/10000', 'learning_rate': '0.00001000', 'loss': '-2.608e-08', 'grad_norm': '0.00441514', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.53131104', 'grpo_loop_time': '416.83056641', 'ref_model_time': '112.96421051', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85547763', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.18554688', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.608e-08', 'kl_loss': '0.00892208', 'kl_moving': '0.00892208', 'kl_stepwise': '-1.930e-08', 'gathered_avg_reward_mean': '9.18420410', 'gathered_avg_reward_std': '1.07570803', 'gathered_video_score2_mq_reward_mean': '3.18139648', 'gathered_video_score2_mq_reward_std': '0.29144940', 'gathered_video_score2_overall_reward_mean': '9.18554688', 'gathered_video_score2_overall_reward_std': '1.07358468', 'gathered_video_score2_ta_reward_mean': '2.90661621', 'gathered_video_score2_ta_reward_std': '0.72955745', 'gathered_video_score2_vq_reward_mean': '3.09619141', 'gathered_video_score2_vq_reward_std': '0.44468001', 'gathered_ori_avg_reward_mean': '9.18420410', 'gathered_ori_avg_reward_std': '1.07570815'} +2026-06-27 13:30:47.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 13:30:47.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:53.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_0/index_5099_seed_25766_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 13:32:09.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:32:09.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:21.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_0/index_5099_seed_25767_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:33:46.264 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 13:33:47.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.082 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 289 train_steps 7560 {'step': '289/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00482277', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.27355957', 'grpo_loop_time': '418.56396484', 'ref_model_time': '111.23536682', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89399117', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.54003906', 'token_credit_video_advantage_mean': '-3.027e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.415e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.191e-09', 'kl_loss': '0.00949458', 'kl_moving': '0.00949458', 'kl_stepwise': '-1.932e-08', 'gathered_avg_reward_mean': '9.54052734', 'gathered_avg_reward_std': '1.07655895', 'gathered_video_score2_mq_reward_mean': '3.24023438', 'gathered_video_score2_mq_reward_std': '0.37545240', 'gathered_video_score2_overall_reward_mean': '9.54003906', 'gathered_video_score2_overall_reward_std': '1.07633269', 'gathered_video_score2_ta_reward_mean': '3.20068359', 'gathered_video_score2_ta_reward_std': '0.57181031', 'gathered_video_score2_vq_reward_mean': '3.09960938', 'gathered_video_score2_vq_reward_std': '0.46711549', 'gathered_ori_avg_reward_mean': '9.54052734', 'gathered_ori_avg_reward_std': '1.07655895'} +2026-06-27 13:44:38.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 13:44:38.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:44.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_0/index_7114_seed_5335_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 13:45:58.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:58.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:09.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_0/index_7114_seed_5336_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:47:30.640 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 13:47:31.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:19.861 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 290 train_steps 7600 {'step': '290/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00449820', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.64364624', 'grpo_loop_time': '417.86550903', 'ref_model_time': '111.49210358', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86997187', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.35058594', 'token_credit_video_advantage_mean': '7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.351e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00957983', 'kl_moving': '0.00957983', 'kl_stepwise': '-1.907e-08', 'gathered_avg_reward_mean': '9.34985352', 'gathered_avg_reward_std': '1.03883147', 'gathered_video_score2_mq_reward_mean': '3.21545410', 'gathered_video_score2_mq_reward_std': '0.38149688', 'gathered_video_score2_overall_reward_mean': '9.35058594', 'gathered_video_score2_overall_reward_std': '1.03608406', 'gathered_video_score2_ta_reward_mean': '3.07116699', 'gathered_video_score2_ta_reward_std': '0.55839247', 'gathered_video_score2_vq_reward_mean': '3.06323242', 'gathered_video_score2_vq_reward_std': '0.39638749', 'gathered_ori_avg_reward_mean': '9.34985352', 'gathered_ori_avg_reward_std': '1.03883159'} +2026-06-27 13:58:55.331 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 13:58:57.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:03.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_0/index_7118_seed_24216_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 14:00:17.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:17.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:30.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_0/index_7118_seed_24217_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:01:50.839 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 14:01:51.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.100 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 291 train_steps 7640 {'step': '291/10000', 'learning_rate': '0.00001000', 'loss': '-1.397e-08', 'grad_norm': '0.00480714', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.30868530', 'grpo_loop_time': '419.22827148', 'ref_model_time': '111.79167175', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86530185', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.33886719', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.426e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.397e-08', 'kl_loss': '0.00802954', 'kl_moving': '0.00802954', 'kl_stepwise': '-2.001e-08', 'gathered_avg_reward_mean': '9.34143066', 'gathered_avg_reward_std': '0.90652955', 'gathered_video_score2_mq_reward_mean': '3.16271973', 'gathered_video_score2_mq_reward_std': '0.33560392', 'gathered_video_score2_overall_reward_mean': '9.33886719', 'gathered_video_score2_overall_reward_std': '0.90696055', 'gathered_video_score2_ta_reward_mean': '3.07910156', 'gathered_video_score2_ta_reward_std': '0.49330908', 'gathered_video_score2_vq_reward_mean': '3.09960938', 'gathered_video_score2_vq_reward_std': '0.46459338', 'gathered_ori_avg_reward_mean': '9.34143066', 'gathered_ori_avg_reward_std': '0.90652955'} +2026-06-27 14:12:43.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 14:12:43.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:49.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_0/index_1837_seed_29237_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 14:14:07.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:14:07.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:12.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_0/index_1837_seed_29238_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:15:31.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 14:15:32.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.060 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 292 train_steps 7680 {'step': '292/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00510217', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '281.68676758', 'grpo_loop_time': '417.24169922', 'ref_model_time': '111.73796082', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85349798', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.05957031', 'token_credit_video_advantage_mean': '-2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.487e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00814667', 'kl_moving': '0.00814668', 'kl_stepwise': '-2.043e-08', 'gathered_avg_reward_mean': '9.05712891', 'gathered_avg_reward_std': '0.66503394', 'gathered_video_score2_mq_reward_mean': '3.12377930', 'gathered_video_score2_mq_reward_std': '0.21463600', 'gathered_video_score2_overall_reward_mean': '9.05957031', 'gathered_video_score2_overall_reward_std': '0.66320533', 'gathered_video_score2_ta_reward_mean': '3.01953125', 'gathered_video_score2_ta_reward_std': '0.48644969', 'gathered_video_score2_vq_reward_mean': '2.91381836', 'gathered_video_score2_vq_reward_std': '0.46452352', 'gathered_ori_avg_reward_mean': '9.05712891', 'gathered_ori_avg_reward_std': '0.66503388'} +2026-06-27 14:26:20.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 14:26:20.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:26.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_0/index_5354_seed_855_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 14:27:41.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:41.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:28:50.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_0/index_5354_seed_856_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:29:11.439 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 14:29:12.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.682 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 293 train_steps 7720 {'step': '293/10000', 'learning_rate': '0.00001000', 'loss': '6.985e-09', 'grad_norm': '0.00500918', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.10049438', 'grpo_loop_time': '418.96875000', 'ref_model_time': '113.04067993', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87692219', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.72509766', 'token_credit_video_advantage_mean': '-2.328e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.451e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.260e-09', 'kl_loss': '0.00828988', 'kl_moving': '0.00828988', 'kl_stepwise': '-1.959e-08', 'gathered_avg_reward_mean': '9.72192383', 'gathered_avg_reward_std': '1.21083093', 'gathered_video_score2_mq_reward_mean': '3.27294922', 'gathered_video_score2_mq_reward_std': '0.40203044', 'gathered_video_score2_overall_reward_mean': '9.72509766', 'gathered_video_score2_overall_reward_std': '1.20776021', 'gathered_video_score2_ta_reward_mean': '3.28759766', 'gathered_video_score2_ta_reward_std': '0.55454111', 'gathered_video_score2_vq_reward_mean': '3.16137695', 'gathered_video_score2_vq_reward_std': '0.44654483', 'gathered_ori_avg_reward_mean': '9.72192383', 'gathered_ori_avg_reward_std': '1.21083093'} +2026-06-27 14:40:05.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 14:40:05.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_0/index_9523_seed_1489_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 14:41:26.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:26.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:37.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_0/index_9523_seed_1490_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:42:57.387 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 14:42:58.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.005 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 294 train_steps 7760 {'step': '294/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-10', 'grad_norm': '0.00497449', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.18026733', 'grpo_loop_time': '417.47369385', 'ref_model_time': '112.89431763', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87252361', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.48828125', 'token_credit_video_advantage_mean': '-1.141e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-10', 'kl_loss': '0.00857672', 'kl_moving': '0.00857672', 'kl_stepwise': '-1.951e-08', 'gathered_avg_reward_mean': '9.48742676', 'gathered_avg_reward_std': '0.70959425', 'gathered_video_score2_mq_reward_mean': '3.32080078', 'gathered_video_score2_mq_reward_std': '0.32679662', 'gathered_video_score2_overall_reward_mean': '9.48828125', 'gathered_video_score2_overall_reward_std': '0.70873415', 'gathered_video_score2_ta_reward_mean': '2.89343262', 'gathered_video_score2_ta_reward_std': '0.51005810', 'gathered_video_score2_vq_reward_mean': '3.27319336', 'gathered_video_score2_vq_reward_std': '0.34070688', 'gathered_ori_avg_reward_mean': '9.48742676', 'gathered_ori_avg_reward_std': '0.70959431'} +2026-06-27 14:53:47.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 14:53:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:53.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_0/index_4997_seed_27848_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 14:55:07.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:55:07.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:18.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_0/index_4997_seed_27849_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:56:38.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 14:56:39.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.721 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 295 train_steps 7800 {'step': '295/10000', 'learning_rate': '0.00001000', 'loss': '-1.211e-08', 'grad_norm': '0.01041261', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.81634521', 'grpo_loop_time': '416.65631104', 'ref_model_time': '111.56227112', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86250782', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.22167969', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.469e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.211e-08', 'kl_loss': '0.00865839', 'kl_moving': '0.00865839', 'kl_stepwise': '-1.961e-08', 'gathered_avg_reward_mean': '9.22058105', 'gathered_avg_reward_std': '1.04488504', 'gathered_video_score2_mq_reward_mean': '3.14306641', 'gathered_video_score2_mq_reward_std': '0.31768426', 'gathered_video_score2_overall_reward_mean': '9.22167969', 'gathered_video_score2_overall_reward_std': '1.04250371', 'gathered_video_score2_ta_reward_mean': '2.99255371', 'gathered_video_score2_ta_reward_std': '0.70294791', 'gathered_video_score2_vq_reward_mean': '3.08496094', 'gathered_video_score2_vq_reward_std': '0.45089382', 'gathered_ori_avg_reward_mean': '9.22058105', 'gathered_ori_avg_reward_std': '1.04488516'} +2026-06-27 15:07:26.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 15:07:26.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:26.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_0/index_8863_seed_2844_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 15:08:47.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:47.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:09:57.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_0/index_8863_seed_2845_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:10:17.228 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 15:10:18.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.552 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 296 train_steps 7840 {'step': '296/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00705694', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.65301514', 'grpo_loop_time': '417.03570557', 'ref_model_time': '112.25419617', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84910363', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.33496094', 'token_credit_video_advantage_mean': '-3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.391e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00933992', 'kl_moving': '0.00933992', 'kl_stepwise': '-1.836e-08', 'gathered_avg_reward_mean': '9.33337402', 'gathered_avg_reward_std': '1.19299090', 'gathered_video_score2_mq_reward_mean': '3.20727539', 'gathered_video_score2_mq_reward_std': '0.35551426', 'gathered_video_score2_overall_reward_mean': '9.33496094', 'gathered_video_score2_overall_reward_std': '1.19249403', 'gathered_video_score2_ta_reward_mean': '3.23498535', 'gathered_video_score2_ta_reward_std': '0.56924182', 'gathered_video_score2_vq_reward_mean': '2.89111328', 'gathered_video_score2_vq_reward_std': '0.49187908', 'gathered_ori_avg_reward_mean': '9.33337402', 'gathered_ori_avg_reward_std': '1.19299102'} +2026-06-27 15:21:06.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 15:21:06.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:12.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_0/index_6899_seed_22925_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 15:22:23.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:23.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:37.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_0/index_6899_seed_22926_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:23:57.336 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 15:23:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.769 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 297 train_steps 7880 {'step': '297/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00515617', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.37646484', 'grpo_loop_time': '417.85839844', 'ref_model_time': '113.01938629', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88055789', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.21630859', 'token_credit_video_advantage_mean': '-2.095e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.00909192', 'kl_moving': '0.00909192', 'kl_stepwise': '-1.916e-08', 'gathered_avg_reward_mean': '9.21850586', 'gathered_avg_reward_std': '1.07615757', 'gathered_video_score2_mq_reward_mean': '3.12963867', 'gathered_video_score2_mq_reward_std': '0.45093447', 'gathered_video_score2_overall_reward_mean': '9.21630859', 'gathered_video_score2_overall_reward_std': '1.07547939', 'gathered_video_score2_ta_reward_mean': '3.11645508', 'gathered_video_score2_ta_reward_std': '0.58023220', 'gathered_video_score2_vq_reward_mean': '2.97241211', 'gathered_video_score2_vq_reward_std': '0.50711334', 'gathered_ori_avg_reward_mean': '9.21850586', 'gathered_ori_avg_reward_std': '1.07615769'} +2026-06-27 15:34:47.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 15:34:47.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:47.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:54.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_0/index_8511_seed_24643_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 15:36:09.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:36:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:20.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_0/index_8511_seed_24644_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:37:43.460 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 15:37:44.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:31.989 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 298 train_steps 7920 {'step': '298/10000', 'learning_rate': '0.00001000', 'loss': '-1.071e-08', 'grad_norm': '0.00483626', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.25793457', 'grpo_loop_time': '416.90881348', 'ref_model_time': '111.99671936', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86555195', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.58300781', 'token_credit_video_advantage_mean': '8.848e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.417e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.257e-08', 'kl_loss': '0.00902689', 'kl_moving': '0.00902689', 'kl_stepwise': '-1.978e-08', 'gathered_avg_reward_mean': '9.58093262', 'gathered_avg_reward_std': '1.18346679', 'gathered_video_score2_mq_reward_mean': '3.26586914', 'gathered_video_score2_mq_reward_std': '0.38051859', 'gathered_video_score2_overall_reward_mean': '9.58300781', 'gathered_video_score2_overall_reward_std': '1.18786252', 'gathered_video_score2_ta_reward_mean': '3.16076660', 'gathered_video_score2_ta_reward_std': '0.57030213', 'gathered_video_score2_vq_reward_mean': '3.15429688', 'gathered_video_score2_vq_reward_std': '0.48904237', 'gathered_ori_avg_reward_mean': '9.58093262', 'gathered_ori_avg_reward_std': '1.18346679'} +2026-06-27 15:48:32.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 15:48:32.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:38.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_0/index_8465_seed_17968_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 15:49:53.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:53.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:03.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_0/index_8465_seed_17969_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:51:25.031 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 15:51:25.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.091 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 299 train_steps 7960 {'step': '299/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00676514', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.30572510', 'grpo_loop_time': '417.27322388', 'ref_model_time': '111.68859863', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88975388', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.40136719', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.474e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.122e-09', 'kl_loss': '0.01048180', 'kl_moving': '0.01048180', 'kl_stepwise': '-2.046e-08', 'gathered_avg_reward_mean': '9.40319824', 'gathered_avg_reward_std': '1.04515791', 'gathered_video_score2_mq_reward_mean': '3.24047852', 'gathered_video_score2_mq_reward_std': '0.34268987', 'gathered_video_score2_overall_reward_mean': '9.40136719', 'gathered_video_score2_overall_reward_std': '1.04447043', 'gathered_video_score2_ta_reward_mean': '3.07385254', 'gathered_video_score2_ta_reward_std': '0.55733907', 'gathered_video_score2_vq_reward_mean': '3.08886719', 'gathered_video_score2_vq_reward_std': '0.37610337', 'gathered_ori_avg_reward_mean': '9.40319824', 'gathered_ori_avg_reward_std': '1.04515791'} +2026-06-27 16:02:14.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 16:02:14.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:20.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_0/index_3979_seed_22078_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 16:03:34.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:34.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:04:45.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_0/index_3979_seed_22079_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:05:04.329 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 16:05:05.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:15:52.199 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 300 train_steps 8000 {'step': '300/10000', 'learning_rate': '0.00001000', 'loss': '-5.122e-09', 'grad_norm': '0.00606689', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.19177246', 'grpo_loop_time': '417.07644653', 'ref_model_time': '111.36447906', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85693431', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.85253906', 'token_credit_video_advantage_mean': '-5.821e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.427e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.122e-09', 'kl_loss': '0.01060654', 'kl_moving': '0.01060655', 'kl_stepwise': '-1.950e-08', 'gathered_avg_reward_mean': '8.85351562', 'gathered_avg_reward_std': '1.01159430', 'gathered_video_score2_mq_reward_mean': '3.05566406', 'gathered_video_score2_mq_reward_std': '0.25798815', 'gathered_video_score2_overall_reward_mean': '8.85253906', 'gathered_video_score2_overall_reward_std': '1.01171827', 'gathered_video_score2_ta_reward_mean': '2.96179199', 'gathered_video_score2_ta_reward_std': '0.62886095', 'gathered_video_score2_vq_reward_mean': '2.83605957', 'gathered_video_score2_vq_reward_std': '0.46080816', 'gathered_ori_avg_reward_mean': '8.85351562', 'gathered_ori_avg_reward_std': '1.01159430'} +2026-06-27 16:17:03.777 | INFO | __main__:save_checkpoint:1967 - Checkpoint saved at step 300 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-300 +2026-06-27 16:17:03.917 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 300: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-300 +2026-06-27 16:17:06.559 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-275 +2026-06-27 16:17:41.942 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 16:17:44.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:44.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_0/index_9074_seed_29894_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 16:19:04.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:19:04.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:17.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_0/index_9074_seed_29895_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:20:37.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 16:20:38.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.747 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 301 train_steps 8040 {'step': '301/10000', 'learning_rate': '0.00001000', 'loss': '-6.054e-09', 'grad_norm': '0.00860120', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.34878540', 'grpo_loop_time': '417.18621826', 'ref_model_time': '111.32435608', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88899601', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.36572266', 'token_credit_video_advantage_mean': '4.424e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.428e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.779e-09', 'kl_loss': '0.00824339', 'kl_moving': '0.00824339', 'kl_stepwise': '-1.969e-08', 'gathered_avg_reward_mean': '9.36853027', 'gathered_avg_reward_std': '1.03498447', 'gathered_video_score2_mq_reward_mean': '3.20678711', 'gathered_video_score2_mq_reward_std': '0.34678549', 'gathered_video_score2_overall_reward_mean': '9.36572266', 'gathered_video_score2_overall_reward_std': '1.03727937', 'gathered_video_score2_ta_reward_mean': '3.02709961', 'gathered_video_score2_ta_reward_std': '0.57026565', 'gathered_video_score2_vq_reward_mean': '3.13464355', 'gathered_video_score2_vq_reward_std': '0.43308702', 'gathered_ori_avg_reward_mean': '9.36853027', 'gathered_ori_avg_reward_std': '1.03498447'} +2026-06-27 16:31:25.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 16:31:26.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:32.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_0/index_2555_seed_28027_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 16:32:48.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:48.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:33:58.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_0/index_2555_seed_28028_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:34:17.637 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 16:34:18.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.094 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 302 train_steps 8080 {'step': '302/10000', 'learning_rate': '0.00001000', 'loss': '6.985e-09', 'grad_norm': '0.00510663', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.08700562', 'grpo_loop_time': '417.40994263', 'ref_model_time': '111.43737793', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85588342', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.54980469', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.361e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.260e-09', 'kl_loss': '0.00870703', 'kl_moving': '0.00870703', 'kl_stepwise': '-1.873e-08', 'gathered_avg_reward_mean': '9.54870605', 'gathered_avg_reward_std': '1.02201521', 'gathered_video_score2_mq_reward_mean': '3.27978516', 'gathered_video_score2_mq_reward_std': '0.36841881', 'gathered_video_score2_overall_reward_mean': '9.54980469', 'gathered_video_score2_overall_reward_std': '1.02310526', 'gathered_video_score2_ta_reward_mean': '3.14489746', 'gathered_video_score2_ta_reward_std': '0.52571535', 'gathered_video_score2_vq_reward_mean': '3.12402344', 'gathered_video_score2_vq_reward_std': '0.43458566', 'gathered_ori_avg_reward_mean': '9.54870605', 'gathered_ori_avg_reward_std': '1.02201521'} +2026-06-27 16:45:07.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 16:45:07.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:13.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_0/index_3446_seed_14871_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 16:46:30.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:30.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:39.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_0/index_3446_seed_14872_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 16:47:58.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:47:58.371 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 16:47:59.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.431 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 303 train_steps 8120 {'step': '303/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-10', 'grad_norm': '0.00472258', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.70281982', 'grpo_loop_time': '416.77218628', 'ref_model_time': '112.88421631', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89142108', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.50732422', 'token_credit_video_advantage_mean': '-1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.434e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-10', 'kl_loss': '0.00790912', 'kl_moving': '0.00790912', 'kl_stepwise': '-1.848e-08', 'gathered_avg_reward_mean': '9.50390625', 'gathered_avg_reward_std': '1.26261413', 'gathered_video_score2_mq_reward_mean': '3.28967285', 'gathered_video_score2_mq_reward_std': '0.46004969', 'gathered_video_score2_overall_reward_mean': '9.50732422', 'gathered_video_score2_overall_reward_std': '1.26319623', 'gathered_video_score2_ta_reward_mean': '3.07031250', 'gathered_video_score2_ta_reward_std': '0.64676905', 'gathered_video_score2_vq_reward_mean': '3.14392090', 'gathered_video_score2_vq_reward_std': '0.52930194', 'gathered_ori_avg_reward_mean': '9.50390625', 'gathered_ori_avg_reward_std': '1.26261413'} +2026-06-27 16:58:47.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 16:58:47.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:53.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_0/index_454_seed_7519_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 17:00:11.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:00:11.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:21.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_0/index_454_seed_7520_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:01:41.128 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 17:01:42.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.117 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 304 train_steps 8160 {'step': '304/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00821414', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.07907104', 'grpo_loop_time': '418.46057129', 'ref_model_time': '111.64058685', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86380333', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14355469', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.471e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00903467', 'kl_moving': '0.00903467', 'kl_stepwise': '-1.972e-08', 'gathered_avg_reward_mean': '9.14343262', 'gathered_avg_reward_std': '1.14548028', 'gathered_video_score2_mq_reward_mean': '3.17016602', 'gathered_video_score2_mq_reward_std': '0.36717832', 'gathered_video_score2_overall_reward_mean': '9.14355469', 'gathered_video_score2_overall_reward_std': '1.14530718', 'gathered_video_score2_ta_reward_mean': '3.07714844', 'gathered_video_score2_ta_reward_std': '0.64719278', 'gathered_video_score2_vq_reward_mean': '2.89611816', 'gathered_video_score2_vq_reward_std': '0.47073287', 'gathered_ori_avg_reward_mean': '9.14343262', 'gathered_ori_avg_reward_std': '1.14548028'} +2026-06-27 17:12:31.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 17:12:31.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:37.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_0/index_9739_seed_32850_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 17:13:53.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:53.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:02.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_0/index_9739_seed_32851_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:15:21.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 17:15:22.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.914 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 305 train_steps 8200 {'step': '305/10000', 'learning_rate': '0.00001000', 'loss': '1.118e-08', 'grad_norm': '0.00582583', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.22235107', 'grpo_loop_time': '419.85769653', 'ref_model_time': '112.07611084', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85751379', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.27099609', 'token_credit_video_advantage_mean': '-7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.369e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00837222', 'kl_moving': '0.00837222', 'kl_stepwise': '-1.900e-08', 'gathered_avg_reward_mean': '9.27697754', 'gathered_avg_reward_std': '1.32481778', 'gathered_video_score2_mq_reward_mean': '3.22290039', 'gathered_video_score2_mq_reward_std': '0.39682811', 'gathered_video_score2_overall_reward_mean': '9.27099609', 'gathered_video_score2_overall_reward_std': '1.32303774', 'gathered_video_score2_ta_reward_mean': '2.97351074', 'gathered_video_score2_ta_reward_std': '0.69236648', 'gathered_video_score2_vq_reward_mean': '3.08056641', 'gathered_video_score2_vq_reward_std': '0.54105771', 'gathered_ori_avg_reward_mean': '9.27697754', 'gathered_ori_avg_reward_std': '1.32481778'} +2026-06-27 17:26:14.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 17:26:15.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_0/index_128_seed_19083_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 17:27:40.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:40.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:28:46.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_0/index_128_seed_19084_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:29:08.252 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 17:29:09.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.678 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 306 train_steps 8240 {'step': '306/10000', 'learning_rate': '0.00001000', 'loss': '1.397e-08', 'grad_norm': '0.00787338', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.04699707', 'grpo_loop_time': '419.87646484', 'ref_model_time': '111.11523438', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86248398', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.93212891', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.412e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.024e-08', 'kl_loss': '0.00921141', 'kl_moving': '0.00921142', 'kl_stepwise': '-1.964e-08', 'gathered_avg_reward_mean': '8.93432617', 'gathered_avg_reward_std': '1.51052320', 'gathered_video_score2_mq_reward_mean': '3.13903809', 'gathered_video_score2_mq_reward_std': '0.51231700', 'gathered_video_score2_overall_reward_mean': '8.93212891', 'gathered_video_score2_overall_reward_std': '1.50832403', 'gathered_video_score2_ta_reward_mean': '2.94250488', 'gathered_video_score2_ta_reward_std': '0.67368567', 'gathered_video_score2_vq_reward_mean': '2.85278320', 'gathered_video_score2_vq_reward_std': '0.57667023', 'gathered_ori_avg_reward_mean': '8.93432617', 'gathered_ori_avg_reward_std': '1.51052332'} +2026-06-27 17:40:00.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 17:40:00.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:00.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:06.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_0/index_3909_seed_11218_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 17:41:21.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:21.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:35.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_0/index_3909_seed_11219_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:42:54.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 17:42:55.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.427 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 307 train_steps 8280 {'step': '307/10000', 'learning_rate': '0.00001000', 'loss': '-2.328e-09', 'grad_norm': '0.00470341', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.37985229', 'grpo_loop_time': '417.64697266', 'ref_model_time': '113.94931030', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88029766', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.65869141', 'token_credit_video_advantage_mean': '2.561e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.328e-09', 'kl_loss': '0.00879624', 'kl_moving': '0.00879624', 'kl_stepwise': '-1.967e-08', 'gathered_avg_reward_mean': '9.65539551', 'gathered_avg_reward_std': '1.03339338', 'gathered_video_score2_mq_reward_mean': '3.32690430', 'gathered_video_score2_mq_reward_std': '0.35593316', 'gathered_video_score2_overall_reward_mean': '9.65869141', 'gathered_video_score2_overall_reward_std': '1.03555131', 'gathered_video_score2_ta_reward_mean': '3.09570312', 'gathered_video_score2_ta_reward_std': '0.62215847', 'gathered_video_score2_vq_reward_mean': '3.23278809', 'gathered_video_score2_vq_reward_std': '0.48454267', 'gathered_ori_avg_reward_mean': '9.65539551', 'gathered_ori_avg_reward_std': '1.03339338'} +2026-06-27 17:53:45.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 17:53:45.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:51.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_0/index_5368_seed_27845_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 17:55:06.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:55:06.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:16.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_0/index_5368_seed_27846_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:56:37.318 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 17:56:38.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.402 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 308 train_steps 8320 {'step': '308/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00581065', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.48999023', 'grpo_loop_time': '417.52581787', 'ref_model_time': '111.04783630', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88533509', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.64208984', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.453e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.00958974', 'kl_moving': '0.00958974', 'kl_stepwise': '-1.915e-08', 'gathered_avg_reward_mean': '9.64013672', 'gathered_avg_reward_std': '1.28415179', 'gathered_video_score2_mq_reward_mean': '3.29833984', 'gathered_video_score2_mq_reward_std': '0.42157397', 'gathered_video_score2_overall_reward_mean': '9.64208984', 'gathered_video_score2_overall_reward_std': '1.28793669', 'gathered_video_score2_ta_reward_mean': '3.21203613', 'gathered_video_score2_ta_reward_std': '0.62384826', 'gathered_video_score2_vq_reward_mean': '3.12976074', 'gathered_video_score2_vq_reward_std': '0.55080140', 'gathered_ori_avg_reward_mean': '9.64013672', 'gathered_ori_avg_reward_std': '1.28415179'} +2026-06-27 18:07:25.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 18:07:25.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:31.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_0/index_1780_seed_27246_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 18:08:44.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:44.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:01.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_0/index_1780_seed_27247_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:10:22.998 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 18:10:24.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.471 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 309 train_steps 8360 {'step': '309/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00775478', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.79534912', 'grpo_loop_time': '417.98980713', 'ref_model_time': '111.33271027', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87199062', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.16796875', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.381e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.01073350', 'kl_moving': '0.01073351', 'kl_stepwise': '-1.969e-08', 'gathered_avg_reward_mean': '9.16711426', 'gathered_avg_reward_std': '1.10544598', 'gathered_video_score2_mq_reward_mean': '3.22192383', 'gathered_video_score2_mq_reward_std': '0.36197871', 'gathered_video_score2_overall_reward_mean': '9.16796875', 'gathered_video_score2_overall_reward_std': '1.11068773', 'gathered_video_score2_ta_reward_mean': '3.03051758', 'gathered_video_score2_ta_reward_std': '0.51553941', 'gathered_video_score2_vq_reward_mean': '2.91467285', 'gathered_video_score2_vq_reward_std': '0.50074011', 'gathered_ori_avg_reward_mean': '9.16711426', 'gathered_ori_avg_reward_std': '1.10544598'} +2026-06-27 18:21:14.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 18:21:14.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_0/index_6573_seed_18771_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 18:22:36.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:36.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:23:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_0/index_6573_seed_18772_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:24:06.987 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 18:24:08.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:01.167 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 310 train_steps 8400 {'step': '310/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-10', 'grad_norm': '0.00518118', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.46362305', 'grpo_loop_time': '420.61343384', 'ref_model_time': '111.11631012', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86479330', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.87841797', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.418e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00955170', 'kl_moving': '0.00955170', 'kl_stepwise': '-1.915e-08', 'gathered_avg_reward_mean': '8.87597656', 'gathered_avg_reward_std': '0.93956536', 'gathered_video_score2_mq_reward_mean': '3.05541992', 'gathered_video_score2_mq_reward_std': '0.19564287', 'gathered_video_score2_overall_reward_mean': '8.87841797', 'gathered_video_score2_overall_reward_std': '0.93912810', 'gathered_video_score2_ta_reward_mean': '2.91442871', 'gathered_video_score2_ta_reward_std': '0.55034029', 'gathered_video_score2_vq_reward_mean': '2.90612793', 'gathered_video_score2_vq_reward_std': '0.48293656', 'gathered_ori_avg_reward_mean': '8.87597656', 'gathered_ori_avg_reward_std': '0.93956530'} +2026-06-27 18:35:36.799 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 18:35:38.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:44.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_0/index_6833_seed_11968_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 18:37:02.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:37:02.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:09.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_0/index_6833_seed_11969_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:38:31.825 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 18:38:32.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:20.931 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 311 train_steps 8440 {'step': '311/10000', 'learning_rate': '0.00001000', 'loss': '-2.235e-08', 'grad_norm': '0.00890632', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.43347168', 'grpo_loop_time': '418.09994507', 'ref_model_time': '111.49071503', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87776852', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.16943359', 'token_credit_video_advantage_mean': '4.424e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.386e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.235e-08', 'kl_loss': '0.00851177', 'kl_moving': '0.00851177', 'kl_stepwise': '-1.930e-08', 'gathered_avg_reward_mean': '9.16845703', 'gathered_avg_reward_std': '0.92386520', 'gathered_video_score2_mq_reward_mean': '3.16796875', 'gathered_video_score2_mq_reward_std': '0.31038100', 'gathered_video_score2_overall_reward_mean': '9.16943359', 'gathered_video_score2_overall_reward_std': '0.92445028', 'gathered_video_score2_ta_reward_mean': '3.11840820', 'gathered_video_score2_ta_reward_std': '0.44891977', 'gathered_video_score2_vq_reward_mean': '2.88208008', 'gathered_video_score2_vq_reward_std': '0.43657416', 'gathered_ori_avg_reward_mean': '9.16845703', 'gathered_ori_avg_reward_std': '0.92386514'} +2026-06-27 18:49:21.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 18:49:21.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:27.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_0/index_5117_seed_25587_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 18:50:40.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:40.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:51:56.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_0/index_5117_seed_25588_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:52:15.806 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 18:52:16.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.397 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 312 train_steps 8480 {'step': '312/10000', 'learning_rate': '0.00001000', 'loss': '-2.235e-08', 'grad_norm': '0.00553862', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.63601685', 'grpo_loop_time': '418.62719727', 'ref_model_time': '112.18264771', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85553199', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.47412109', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.485e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.235e-08', 'kl_loss': '0.00842362', 'kl_moving': '0.00842363', 'kl_stepwise': '-2.051e-08', 'gathered_avg_reward_mean': '9.46899414', 'gathered_avg_reward_std': '0.90623528', 'gathered_video_score2_mq_reward_mean': '3.15112305', 'gathered_video_score2_mq_reward_std': '0.30703291', 'gathered_video_score2_overall_reward_mean': '9.47412109', 'gathered_video_score2_overall_reward_std': '0.90881497', 'gathered_video_score2_ta_reward_mean': '3.16455078', 'gathered_video_score2_ta_reward_std': '0.47749788', 'gathered_video_score2_vq_reward_mean': '3.15332031', 'gathered_video_score2_vq_reward_std': '0.33969638', 'gathered_ori_avg_reward_mean': '9.46899414', 'gathered_ori_avg_reward_std': '0.90623522'} +2026-06-27 19:03:06.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 19:03:06.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:13.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_0/index_1303_seed_9598_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 19:04:30.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:30.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:36.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_0/index_1303_seed_9599_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:05:54.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 19:05:55.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.821 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 313 train_steps 8520 {'step': '313/10000', 'learning_rate': '0.00001000', 'loss': '-7.334e-09', 'grad_norm': '0.00575073', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.23852539', 'grpo_loop_time': '417.53527832', 'ref_model_time': '114.07104492', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89745265', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.22656250', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.486e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.334e-09', 'kl_loss': '0.00861291', 'kl_moving': '0.00861291', 'kl_stepwise': '-1.916e-08', 'gathered_avg_reward_mean': '9.22167969', 'gathered_avg_reward_std': '0.76266283', 'gathered_video_score2_mq_reward_mean': '3.16235352', 'gathered_video_score2_mq_reward_std': '0.24186623', 'gathered_video_score2_overall_reward_mean': '9.22656250', 'gathered_video_score2_overall_reward_std': '0.76390910', 'gathered_video_score2_ta_reward_mean': '3.04687500', 'gathered_video_score2_ta_reward_std': '0.58209515', 'gathered_video_score2_vq_reward_mean': '3.01245117', 'gathered_video_score2_vq_reward_std': '0.31385425', 'gathered_ori_avg_reward_mean': '9.22167969', 'gathered_ori_avg_reward_std': '0.76266283'} +2026-06-27 19:16:45.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 19:16:45.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:45.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_0/index_2436_seed_30250_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 19:18:10.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:18:10.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:21.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_0/index_2436_seed_30251_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:19:41.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 19:19:42.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.311 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 314 train_steps 8560 {'step': '314/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-10', 'grad_norm': '0.00489313', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.05871582', 'grpo_loop_time': '417.43997192', 'ref_model_time': '111.21575165', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87541175', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.75195312', 'token_credit_video_advantage_mean': '1.001e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.472e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.00879912', 'kl_moving': '0.00879912', 'kl_stepwise': '-1.888e-08', 'gathered_avg_reward_mean': '9.75439453', 'gathered_avg_reward_std': '0.67418456', 'gathered_video_score2_mq_reward_mean': '3.26049805', 'gathered_video_score2_mq_reward_std': '0.27252996', 'gathered_video_score2_overall_reward_mean': '9.75195312', 'gathered_video_score2_overall_reward_std': '0.67205381', 'gathered_video_score2_ta_reward_mean': '3.34912109', 'gathered_video_score2_ta_reward_std': '0.42002833', 'gathered_video_score2_vq_reward_mean': '3.14477539', 'gathered_video_score2_vq_reward_std': '0.40433219', 'gathered_ori_avg_reward_mean': '9.75439453', 'gathered_ori_avg_reward_std': '0.67418450'} +2026-06-27 19:30:30.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 19:30:30.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:36.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_0/index_9778_seed_18904_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 19:31:51.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:51.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:04.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_0/index_9778_seed_18905_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:33:24.327 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 19:33:25.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.910 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 315 train_steps 8600 {'step': '315/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00617523', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.78131104', 'grpo_loop_time': '417.92553711', 'ref_model_time': '111.16878510', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86364681', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.76806641', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00907358', 'kl_moving': '0.00907358', 'kl_stepwise': '-1.937e-08', 'gathered_avg_reward_mean': '9.76354980', 'gathered_avg_reward_std': '1.06726015', 'gathered_video_score2_mq_reward_mean': '3.34741211', 'gathered_video_score2_mq_reward_std': '0.33797237', 'gathered_video_score2_overall_reward_mean': '9.76806641', 'gathered_video_score2_overall_reward_std': '1.07013857', 'gathered_video_score2_ta_reward_mean': '3.33215332', 'gathered_video_score2_ta_reward_std': '0.60638946', 'gathered_video_score2_vq_reward_mean': '3.08398438', 'gathered_video_score2_vq_reward_std': '0.42193377', 'gathered_ori_avg_reward_mean': '9.76354980', 'gathered_ori_avg_reward_std': '1.06726015'} +2026-06-27 19:44:13.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 19:44:14.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_0/index_7346_seed_7520_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 19:45:37.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:37.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:46:44.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_0/index_7346_seed_7521_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 19:47:08.992 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:47:09.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 19:47:10.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.178 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 316 train_steps 8640 {'step': '316/10000', 'learning_rate': '0.00001000', 'loss': '1.676e-08', 'grad_norm': '0.00533182', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.95050049', 'grpo_loop_time': '420.60205078', 'ref_model_time': '111.74071503', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86599886', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.33349609', 'token_credit_video_advantage_mean': '-7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.412e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.676e-08', 'kl_loss': '0.00946031', 'kl_moving': '0.00946031', 'kl_stepwise': '-2.001e-08', 'gathered_avg_reward_mean': '9.33154297', 'gathered_avg_reward_std': '0.90936154', 'gathered_video_score2_mq_reward_mean': '3.25878906', 'gathered_video_score2_mq_reward_std': '0.41006207', 'gathered_video_score2_overall_reward_mean': '9.33349609', 'gathered_video_score2_overall_reward_std': '0.90907520', 'gathered_video_score2_ta_reward_mean': '2.85925293', 'gathered_video_score2_ta_reward_std': '0.57815385', 'gathered_video_score2_vq_reward_mean': '3.21350098', 'gathered_video_score2_vq_reward_std': '0.38880450', 'gathered_ori_avg_reward_mean': '9.33154297', 'gathered_ori_avg_reward_std': '0.90936154'} +2026-06-27 19:58:02.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 19:58:02.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_0/index_1131_seed_17444_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 19:59:26.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:26.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:33.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_0/index_1131_seed_17445_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:00:53.161 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 20:00:54.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.659 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 317 train_steps 8680 {'step': '317/10000', 'learning_rate': '0.00001000', 'loss': '9.779e-09', 'grad_norm': '0.00516817', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.99609375', 'grpo_loop_time': '418.04196167', 'ref_model_time': '111.38304901', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89000130', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.45849609', 'token_credit_video_advantage_mean': '4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.420e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.779e-09', 'kl_loss': '0.00949685', 'kl_moving': '0.00949686', 'kl_stepwise': '-1.979e-08', 'gathered_avg_reward_mean': '9.46289062', 'gathered_avg_reward_std': '1.43772352', 'gathered_video_score2_mq_reward_mean': '3.23535156', 'gathered_video_score2_mq_reward_std': '0.47957769', 'gathered_video_score2_overall_reward_mean': '9.45849609', 'gathered_video_score2_overall_reward_std': '1.43156528', 'gathered_video_score2_ta_reward_mean': '3.15161133', 'gathered_video_score2_ta_reward_std': '0.71568072', 'gathered_video_score2_vq_reward_mean': '3.07592773', 'gathered_video_score2_vq_reward_std': '0.51980925', 'gathered_ori_avg_reward_mean': '9.46289062', 'gathered_ori_avg_reward_std': '1.43772352'} +2026-06-27 20:11:43.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 20:11:43.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:43.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:50.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_0/index_8909_seed_24565_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 20:13:05.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:13:05.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:16.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_0/index_8909_seed_24566_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:14:38.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 20:14:39.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:25.992 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 318 train_steps 8720 {'step': '318/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00726448', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.39590454', 'grpo_loop_time': '416.63088989', 'ref_model_time': '111.26176453', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88042921', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.22851562', 'token_credit_video_advantage_mean': '3.958e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.458e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00909352', 'kl_moving': '0.00909352', 'kl_stepwise': '-1.877e-08', 'gathered_avg_reward_mean': '9.22961426', 'gathered_avg_reward_std': '0.71441752', 'gathered_video_score2_mq_reward_mean': '3.20336914', 'gathered_video_score2_mq_reward_std': '0.31968656', 'gathered_video_score2_overall_reward_mean': '9.22851562', 'gathered_video_score2_overall_reward_std': '0.71808892', 'gathered_video_score2_ta_reward_mean': '2.92895508', 'gathered_video_score2_ta_reward_std': '0.42745271', 'gathered_video_score2_vq_reward_mean': '3.09729004', 'gathered_video_score2_vq_reward_std': '0.38899544', 'gathered_ori_avg_reward_mean': '9.22961426', 'gathered_ori_avg_reward_std': '0.71441752'} +2026-06-27 20:25:26.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 20:25:26.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_0/index_919_seed_561_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 20:26:47.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:47.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:27:59.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_0/index_919_seed_562_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:28:18.151 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 20:28:19.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.606 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 319 train_steps 8760 {'step': '319/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00738548', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.34759521', 'grpo_loop_time': '417.01278687', 'ref_model_time': '111.41220856', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84659266', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.37060547', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.365e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.01044494', 'kl_moving': '0.01044494', 'kl_stepwise': '-1.849e-08', 'gathered_avg_reward_mean': '9.36816406', 'gathered_avg_reward_std': '1.30839801', 'gathered_video_score2_mq_reward_mean': '3.14648438', 'gathered_video_score2_mq_reward_std': '0.42196089', 'gathered_video_score2_overall_reward_mean': '9.37060547', 'gathered_video_score2_overall_reward_std': '1.31194031', 'gathered_video_score2_ta_reward_mean': '3.24145508', 'gathered_video_score2_ta_reward_std': '0.53914690', 'gathered_video_score2_vq_reward_mean': '2.98022461', 'gathered_video_score2_vq_reward_std': '0.60211229', 'gathered_ori_avg_reward_mean': '9.36816406', 'gathered_ori_avg_reward_std': '1.30839801'} +2026-06-27 20:39:06.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 20:39:06.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:13.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_0/index_2135_seed_16974_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 20:40:30.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:30.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:41:39.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_0/index_2135_seed_16975_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:42:03.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 20:42:04.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:52:52.155 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 320 train_steps 8800 {'step': '320/10000', 'learning_rate': '0.00001000', 'loss': '1.118e-08', 'grad_norm': '0.00743574', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.53540039', 'grpo_loop_time': '417.69476318', 'ref_model_time': '110.90583038', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87930894', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.70751953', 'token_credit_video_advantage_mean': '-6.752e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.441e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.122e-09', 'kl_loss': '0.01094557', 'kl_moving': '0.01094557', 'kl_stepwise': '-1.968e-08', 'gathered_avg_reward_mean': '9.70629883', 'gathered_avg_reward_std': '1.19154763', 'gathered_video_score2_mq_reward_mean': '3.25512695', 'gathered_video_score2_mq_reward_std': '0.40232530', 'gathered_video_score2_overall_reward_mean': '9.70751953', 'gathered_video_score2_overall_reward_std': '1.19011033', 'gathered_video_score2_ta_reward_mean': '3.43115234', 'gathered_video_score2_ta_reward_std': '0.43954182', 'gathered_video_score2_vq_reward_mean': '3.02001953', 'gathered_video_score2_vq_reward_std': '0.57195872', 'gathered_ori_avg_reward_mean': '9.70629883', 'gathered_ori_avg_reward_std': '1.19154775'} +2026-06-27 20:53:27.434 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 20:53:29.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:35.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_0/index_9566_seed_29737_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 20:54:53.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:53.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:55:59.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_0/index_9566_seed_29738_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:56:20.460 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 20:56:21.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.479 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 321 train_steps 8840 {'step': '321/10000', 'learning_rate': '0.00001000', 'loss': '2.049e-08', 'grad_norm': '0.00514834', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.49468994', 'grpo_loop_time': '417.61413574', 'ref_model_time': '111.60633850', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86464179', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.09228516', 'token_credit_video_advantage_mean': '6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.445e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.049e-08', 'kl_loss': '0.00813793', 'kl_moving': '0.00813793', 'kl_stepwise': '-1.999e-08', 'gathered_avg_reward_mean': '9.09643555', 'gathered_avg_reward_std': '1.17405498', 'gathered_video_score2_mq_reward_mean': '3.14013672', 'gathered_video_score2_mq_reward_std': '0.39254382', 'gathered_video_score2_overall_reward_mean': '9.09228516', 'gathered_video_score2_overall_reward_std': '1.17943430', 'gathered_video_score2_ta_reward_mean': '2.92822266', 'gathered_video_score2_ta_reward_std': '0.58230627', 'gathered_video_score2_vq_reward_mean': '3.02807617', 'gathered_video_score2_vq_reward_std': '0.44607580', 'gathered_ori_avg_reward_mean': '9.09643555', 'gathered_ori_avg_reward_std': '1.17405498'} +2026-06-27 21:07:10.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 21:07:10.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:17.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_0/index_8632_seed_15629_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 21:08:33.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:33.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:09:41.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_0/index_8632_seed_15630_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:10:06.197 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 21:10:07.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:00.978 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 322 train_steps 8880 {'step': '322/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00649256', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.19946289', 'grpo_loop_time': '421.87750244', 'ref_model_time': '110.86624146', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85675371', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.47460938', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00797511', 'kl_moving': '0.00797511', 'kl_stepwise': '-1.915e-08', 'gathered_avg_reward_mean': '9.47351074', 'gathered_avg_reward_std': '0.94267339', 'gathered_video_score2_mq_reward_mean': '3.24926758', 'gathered_video_score2_mq_reward_std': '0.34019637', 'gathered_video_score2_overall_reward_mean': '9.47460938', 'gathered_video_score2_overall_reward_std': '0.94668150', 'gathered_video_score2_ta_reward_mean': '2.98986816', 'gathered_video_score2_ta_reward_std': '0.59856462', 'gathered_video_score2_vq_reward_mean': '3.23437500', 'gathered_video_score2_vq_reward_std': '0.37768611', 'gathered_ori_avg_reward_mean': '9.47351074', 'gathered_ori_avg_reward_std': '0.94267333'} +2026-06-27 21:21:01.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 21:21:01.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:07.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_0/index_7856_seed_25236_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 21:22:24.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:24.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:32.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_0/index_7856_seed_25237_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:23:51.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 21:23:52.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.139 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 323 train_steps 8920 {'step': '323/10000', 'learning_rate': '0.00001000', 'loss': '2.654e-08', 'grad_norm': '0.00598105', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.90100098', 'grpo_loop_time': '418.84548950', 'ref_model_time': '111.13950348', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86652857', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.58398438', 'token_credit_video_advantage_mean': '2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.427e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.654e-08', 'kl_loss': '0.00874483', 'kl_moving': '0.00874483', 'kl_stepwise': '-1.841e-08', 'gathered_avg_reward_mean': '9.58398438', 'gathered_avg_reward_std': '0.97960174', 'gathered_video_score2_mq_reward_mean': '3.23144531', 'gathered_video_score2_mq_reward_std': '0.35986161', 'gathered_video_score2_overall_reward_mean': '9.58398438', 'gathered_video_score2_overall_reward_std': '0.97741956', 'gathered_video_score2_ta_reward_mean': '3.11840820', 'gathered_video_score2_ta_reward_std': '0.55943006', 'gathered_video_score2_vq_reward_mean': '3.23413086', 'gathered_video_score2_vq_reward_std': '0.31177273', 'gathered_ori_avg_reward_mean': '9.58398438', 'gathered_ori_avg_reward_std': '0.97960174'} +2026-06-27 21:34:42.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 21:34:42.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:48.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_0/index_3360_seed_15333_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 21:36:03.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:36:03.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:13.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_0/index_3360_seed_15334_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:37:33.818 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 21:37:34.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.790 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 324 train_steps 8960 {'step': '324/10000', 'learning_rate': '0.00001000', 'loss': '-3.725e-09', 'grad_norm': '0.00577313', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.72982788', 'grpo_loop_time': '419.09594727', 'ref_model_time': '111.47454834', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88322216', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.40576172', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.438e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.00890151', 'kl_moving': '0.00890151', 'kl_stepwise': '-1.973e-08', 'gathered_avg_reward_mean': '9.40600586', 'gathered_avg_reward_std': '0.99749249', 'gathered_video_score2_mq_reward_mean': '3.17724609', 'gathered_video_score2_mq_reward_std': '0.40813509', 'gathered_video_score2_overall_reward_mean': '9.40576172', 'gathered_video_score2_overall_reward_std': '0.99408484', 'gathered_video_score2_ta_reward_mean': '3.14892578', 'gathered_video_score2_ta_reward_std': '0.44520086', 'gathered_video_score2_vq_reward_mean': '3.07983398', 'gathered_video_score2_vq_reward_std': '0.40758231', 'gathered_ori_avg_reward_mean': '9.40600586', 'gathered_ori_avg_reward_std': '0.99749249'} +2026-06-27 21:48:23.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 21:48:23.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:23.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:30.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_0/index_4144_seed_28232_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 21:49:44.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:44.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:50:55.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_0/index_4144_seed_28233_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 21:51:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:51:15.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 21:51:15.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:02:07.069 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 325 train_steps 9000 {'step': '325/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00619955', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.97723389', 'grpo_loop_time': '419.23617554', 'ref_model_time': '112.03958130', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84200501', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.51318359', 'token_credit_video_advantage_mean': '8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.426e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00912816', 'kl_moving': '0.00912816', 'kl_stepwise': '-1.891e-08', 'gathered_avg_reward_mean': '9.51611328', 'gathered_avg_reward_std': '1.14912724', 'gathered_video_score2_mq_reward_mean': '3.30224609', 'gathered_video_score2_mq_reward_std': '0.48509875', 'gathered_video_score2_overall_reward_mean': '9.51318359', 'gathered_video_score2_overall_reward_std': '1.14616060', 'gathered_video_score2_ta_reward_mean': '3.07592773', 'gathered_video_score2_ta_reward_std': '0.61981565', 'gathered_video_score2_vq_reward_mean': '3.13793945', 'gathered_video_score2_vq_reward_std': '0.42019480', 'gathered_ori_avg_reward_mean': '9.51611328', 'gathered_ori_avg_reward_std': '1.14912724'} +2026-06-27 22:03:21.223 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 325 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-325 +2026-06-27 22:03:21.364 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 325: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-325 +2026-06-27 22:03:21.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 22:03:21.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:27.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_0/index_5012_seed_14984_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 22:04:42.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:42.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:05:55.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_0/index_5012_seed_14985_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:06:16.169 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 22:06:17.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.021 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.700 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 326 train_steps 9040 {'step': '326/10000', 'learning_rate': '0.00001000', 'loss': '2.398e-08', 'grad_norm': '0.00674149', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.60107422', 'grpo_loop_time': '418.36285400', 'ref_model_time': '112.31429291', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85222518', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.32128906', 'token_credit_video_advantage_mean': '-2.095e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.403e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.398e-08', 'kl_loss': '0.00960615', 'kl_moving': '0.00960615', 'kl_stepwise': '-1.966e-08', 'gathered_avg_reward_mean': '9.32299805', 'gathered_avg_reward_std': '0.90326619', 'gathered_video_score2_mq_reward_mean': '3.13061523', 'gathered_video_score2_mq_reward_std': '0.34567958', 'gathered_video_score2_overall_reward_mean': '9.32128906', 'gathered_video_score2_overall_reward_std': '0.90335727', 'gathered_video_score2_ta_reward_mean': '3.20056152', 'gathered_video_score2_ta_reward_std': '0.48375845', 'gathered_video_score2_vq_reward_mean': '2.99182129', 'gathered_video_score2_vq_reward_std': '0.36441728', 'gathered_ori_avg_reward_mean': '9.32299805', 'gathered_ori_avg_reward_std': '0.90326619'} +2026-06-27 22:17:06.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 22:17:06.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:12.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_0/index_7674_seed_13414_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 22:18:26.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:26.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:37.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_0/index_7674_seed_13415_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 22:19:56.152 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:19:56.318 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 22:19:57.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.475 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 327 train_steps 9080 {'step': '327/10000', 'learning_rate': '0.00001000', 'loss': '-1.676e-08', 'grad_norm': '0.00732734', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.95166016', 'grpo_loop_time': '419.36853027', 'ref_model_time': '112.72799683', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85187376', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.06835938', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.489e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.676e-08', 'kl_loss': '0.00993858', 'kl_moving': '0.00993859', 'kl_stepwise': '-1.951e-08', 'gathered_avg_reward_mean': '9.07202148', 'gathered_avg_reward_std': '1.07356036', 'gathered_video_score2_mq_reward_mean': '3.12805176', 'gathered_video_score2_mq_reward_std': '0.44409788', 'gathered_video_score2_overall_reward_mean': '9.06835938', 'gathered_video_score2_overall_reward_std': '1.07182086', 'gathered_video_score2_ta_reward_mean': '2.82287598', 'gathered_video_score2_ta_reward_std': '0.57701057', 'gathered_video_score2_vq_reward_mean': '3.12109375', 'gathered_video_score2_vq_reward_std': '0.43682811', 'gathered_ori_avg_reward_mean': '9.07202148', 'gathered_ori_avg_reward_std': '1.07356024'} +2026-06-27 22:30:47.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 22:30:47.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:53.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_0/index_9774_seed_13154_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 22:32:06.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:32:06.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:22.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_0/index_9774_seed_13155_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:33:40.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 22:33:41.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.156 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 328 train_steps 9120 {'step': '328/10000', 'learning_rate': '0.00001000', 'loss': '1.118e-08', 'grad_norm': '0.00672035', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.41879272', 'grpo_loop_time': '418.06976318', 'ref_model_time': '111.46906281', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83911502', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.83544922', 'token_credit_video_advantage_mean': '-1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.468e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.094e-08', 'kl_loss': '0.00987276', 'kl_moving': '0.00987276', 'kl_stepwise': '-1.951e-08', 'gathered_avg_reward_mean': '8.83386230', 'gathered_avg_reward_std': '1.02457488', 'gathered_video_score2_mq_reward_mean': '3.06335449', 'gathered_video_score2_mq_reward_std': '0.34039366', 'gathered_video_score2_overall_reward_mean': '8.83544922', 'gathered_video_score2_overall_reward_std': '1.02202380', 'gathered_video_score2_ta_reward_mean': '2.93688965', 'gathered_video_score2_ta_reward_std': '0.54210097', 'gathered_video_score2_vq_reward_mean': '2.83361816', 'gathered_video_score2_vq_reward_std': '0.49197951', 'gathered_ori_avg_reward_mean': '8.83386230', 'gathered_ori_avg_reward_std': '1.02457488'} +2026-06-27 22:44:30.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 22:44:30.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:36.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_0/index_1042_seed_14994_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 22:45:49.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:49.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:00.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_0/index_1042_seed_14995_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:47:17.433 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 22:47:18.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.798 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 329 train_steps 9160 {'step': '329/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.00888576', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '280.09600830', 'grpo_loop_time': '418.66925049', 'ref_model_time': '111.10870361', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88251477', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.72949219', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.398e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-09', 'kl_loss': '0.01003093', 'kl_moving': '0.01003093', 'kl_stepwise': '-1.914e-08', 'gathered_avg_reward_mean': '8.72851562', 'gathered_avg_reward_std': '1.35490084', 'gathered_video_score2_mq_reward_mean': '3.03161621', 'gathered_video_score2_mq_reward_std': '0.43103576', 'gathered_video_score2_overall_reward_mean': '8.72949219', 'gathered_video_score2_overall_reward_std': '1.35241139', 'gathered_video_score2_ta_reward_mean': '2.77502441', 'gathered_video_score2_ta_reward_std': '0.70121700', 'gathered_video_score2_vq_reward_mean': '2.92187500', 'gathered_video_score2_vq_reward_std': '0.56681186', 'gathered_ori_avg_reward_mean': '8.72851562', 'gathered_ori_avg_reward_std': '1.35490084'} +2026-06-27 22:58:06.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 22:58:06.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:13.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_0/index_2907_seed_30776_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 22:59:33.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:33.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:00:38.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_0/index_2907_seed_30777_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:01:00.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 23:01:01.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:11:52.229 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 330 train_steps 9200 {'step': '330/10000', 'learning_rate': '0.00001000', 'loss': '1.118e-08', 'grad_norm': '0.00484219', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.22488403', 'grpo_loop_time': '418.71429443', 'ref_model_time': '112.44297791', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88854349', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.70996094', 'token_credit_video_advantage_mean': '1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.470e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.118e-08', 'kl_loss': '0.00956641', 'kl_moving': '0.00956642', 'kl_stepwise': '-1.956e-08', 'gathered_avg_reward_mean': '9.70776367', 'gathered_avg_reward_std': '1.02516806', 'gathered_video_score2_mq_reward_mean': '3.26684570', 'gathered_video_score2_mq_reward_std': '0.41125509', 'gathered_video_score2_overall_reward_mean': '9.70996094', 'gathered_video_score2_overall_reward_std': '1.02579749', 'gathered_video_score2_ta_reward_mean': '3.22607422', 'gathered_video_score2_ta_reward_std': '0.53499919', 'gathered_video_score2_vq_reward_mean': '3.21484375', 'gathered_video_score2_vq_reward_std': '0.38713282', 'gathered_ori_avg_reward_mean': '9.70776367', 'gathered_ori_avg_reward_std': '1.02516806'} +2026-06-27 23:12:27.656 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 23:12:29.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_0/index_7128_seed_21624_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 23:13:50.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:50.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:02.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_0/index_7128_seed_21625_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:15:22.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 23:15:23.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:14.901 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 331 train_steps 9240 {'step': '331/10000', 'learning_rate': '0.00001000', 'loss': '-1.770e-08', 'grad_norm': '0.00637297', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.38745117', 'grpo_loop_time': '418.63775635', 'ref_model_time': '111.28853607', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85685152', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.34375000', 'token_credit_video_advantage_mean': '5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.506e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.770e-08', 'kl_loss': '0.00863580', 'kl_moving': '0.00863580', 'kl_stepwise': '-1.977e-08', 'gathered_avg_reward_mean': '9.34680176', 'gathered_avg_reward_std': '1.00754142', 'gathered_video_score2_mq_reward_mean': '3.22753906', 'gathered_video_score2_mq_reward_std': '0.33264965', 'gathered_video_score2_overall_reward_mean': '9.34375000', 'gathered_video_score2_overall_reward_std': '1.00941610', 'gathered_video_score2_ta_reward_mean': '3.01770020', 'gathered_video_score2_ta_reward_std': '0.57569206', 'gathered_video_score2_vq_reward_mean': '3.10156250', 'gathered_video_score2_vq_reward_std': '0.47908527', 'gathered_ori_avg_reward_mean': '9.34680176', 'gathered_ori_avg_reward_std': '1.00754142'} +2026-06-27 23:26:15.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 23:26:15.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:21.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_0/index_5373_seed_32886_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 23:27:38.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:38.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:28:45.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_0/index_5373_seed_32887_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:29:06.078 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 23:29:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.220 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.059 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 332 train_steps 9280 {'step': '332/10000', 'learning_rate': '0.00001000', 'loss': '-5.122e-09', 'grad_norm': '0.00633106', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.01705933', 'grpo_loop_time': '422.18582153', 'ref_model_time': '111.34696960', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87485039', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.02539062', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.476e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.122e-09', 'kl_loss': '0.00887961', 'kl_moving': '0.00887961', 'kl_stepwise': '-2.016e-08', 'gathered_avg_reward_mean': '9.02563477', 'gathered_avg_reward_std': '1.10167456', 'gathered_video_score2_mq_reward_mean': '3.14086914', 'gathered_video_score2_mq_reward_std': '0.22118641', 'gathered_video_score2_overall_reward_mean': '9.02539062', 'gathered_video_score2_overall_reward_std': '1.10257149', 'gathered_video_score2_ta_reward_mean': '2.97644043', 'gathered_video_score2_ta_reward_std': '0.63518089', 'gathered_video_score2_vq_reward_mean': '2.90832520', 'gathered_video_score2_vq_reward_std': '0.47295526', 'gathered_ori_avg_reward_mean': '9.02563477', 'gathered_ori_avg_reward_std': '1.10167456'} +2026-06-27 23:40:02.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 23:40:02.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:08.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_0/index_2288_seed_21658_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 23:41:21.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:21.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:42:36.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_0/index_2288_seed_21659_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:43:00.928 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 23:43:01.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:50.943 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 333 train_steps 9320 {'step': '333/10000', 'learning_rate': '0.00001000', 'loss': '3.958e-09', 'grad_norm': '0.00566461', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.24514771', 'grpo_loop_time': '419.07351685', 'ref_model_time': '111.69496918', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87127239', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.97460938', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.452e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00854871', 'kl_moving': '0.00854872', 'kl_stepwise': '-2.019e-08', 'gathered_avg_reward_mean': '8.97497559', 'gathered_avg_reward_std': '0.94448102', 'gathered_video_score2_mq_reward_mean': '3.10083008', 'gathered_video_score2_mq_reward_std': '0.26624081', 'gathered_video_score2_overall_reward_mean': '8.97460938', 'gathered_video_score2_overall_reward_std': '0.94540733', 'gathered_video_score2_ta_reward_mean': '2.84191895', 'gathered_video_score2_ta_reward_std': '0.57973069', 'gathered_video_score2_vq_reward_mean': '3.03222656', 'gathered_video_score2_vq_reward_std': '0.30479860', 'gathered_ori_avg_reward_mean': '8.97497559', 'gathered_ori_avg_reward_std': '0.94448102'} +2026-06-27 23:53:50.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-27 23:53:51.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:57.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_0/index_6266_seed_7716_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-27 23:55:12.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:55:12.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:22.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_0/index_6266_seed_7717_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:56:43.681 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-27 23:56:44.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.631 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 334 train_steps 9360 {'step': '334/10000', 'learning_rate': '0.00001000', 'loss': '2.328e-09', 'grad_norm': '0.00716127', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.74487305', 'grpo_loop_time': '417.81036377', 'ref_model_time': '111.43228149', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87266326', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.17089844', 'token_credit_video_advantage_mean': '1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.435e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.397e-09', 'kl_loss': '0.00991734', 'kl_moving': '0.00991734', 'kl_stepwise': '-1.893e-08', 'gathered_avg_reward_mean': '9.17016602', 'gathered_avg_reward_std': '1.01522338', 'gathered_video_score2_mq_reward_mean': '3.13452148', 'gathered_video_score2_mq_reward_std': '0.31441814', 'gathered_video_score2_overall_reward_mean': '9.17089844', 'gathered_video_score2_overall_reward_std': '1.01610518', 'gathered_video_score2_ta_reward_mean': '3.02319336', 'gathered_video_score2_ta_reward_std': '0.66573489', 'gathered_video_score2_vq_reward_mean': '3.01245117', 'gathered_video_score2_vq_reward_std': '0.39922354', 'gathered_ori_avg_reward_mean': '9.17016602', 'gathered_ori_avg_reward_std': '1.01522350'} +2026-06-28 00:07:32.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 00:07:32.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:38.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_0/index_9570_seed_2622_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 00:08:56.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:56.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:08.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_0/index_9570_seed_2623_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:10:30.682 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 00:10:31.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.711 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 335 train_steps 9400 {'step': '335/10000', 'learning_rate': '0.00001000', 'loss': '6.985e-09', 'grad_norm': '0.00696194', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.89230347', 'grpo_loop_time': '419.79589844', 'ref_model_time': '111.23580170', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86258334', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07714844', 'token_credit_video_advantage_mean': '-6.752e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.392e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.985e-09', 'kl_loss': '0.00896608', 'kl_moving': '0.00896608', 'kl_stepwise': '-1.925e-08', 'gathered_avg_reward_mean': '9.08032227', 'gathered_avg_reward_std': '0.93261516', 'gathered_video_score2_mq_reward_mean': '3.15112305', 'gathered_video_score2_mq_reward_std': '0.31874922', 'gathered_video_score2_overall_reward_mean': '9.07714844', 'gathered_video_score2_overall_reward_std': '0.93263662', 'gathered_video_score2_ta_reward_mean': '2.90258789', 'gathered_video_score2_ta_reward_std': '0.59773278', 'gathered_video_score2_vq_reward_mean': '3.02661133', 'gathered_video_score2_vq_reward_std': '0.44174904', 'gathered_ori_avg_reward_mean': '9.08032227', 'gathered_ori_avg_reward_std': '0.93261510'} +2026-06-28 00:21:21.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 00:21:21.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:21.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:28.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_0/index_3095_seed_6549_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 00:22:46.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:46.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:23:52.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_0/index_3095_seed_6550_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:24:11.557 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 00:24:12.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.283 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 336 train_steps 9440 {'step': '336/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00652433', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.69818115', 'grpo_loop_time': '420.67050171', 'ref_model_time': '113.27371216', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86410534', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.39941406', 'token_credit_video_advantage_mean': '-8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.406e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00937341', 'kl_moving': '0.00937341', 'kl_stepwise': '-1.898e-08', 'gathered_avg_reward_mean': '9.39746094', 'gathered_avg_reward_std': '0.91710865', 'gathered_video_score2_mq_reward_mean': '3.19982910', 'gathered_video_score2_mq_reward_std': '0.35113582', 'gathered_video_score2_overall_reward_mean': '9.39941406', 'gathered_video_score2_overall_reward_std': '0.91801810', 'gathered_video_score2_ta_reward_mean': '3.18908691', 'gathered_video_score2_ta_reward_std': '0.53528309', 'gathered_video_score2_vq_reward_mean': '3.00854492', 'gathered_video_score2_vq_reward_std': '0.37733170', 'gathered_ori_avg_reward_mean': '9.39746094', 'gathered_ori_avg_reward_std': '0.91710865'} +2026-06-28 00:35:07.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 00:35:07.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:13.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_0/index_4516_seed_18834_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 00:36:28.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:28.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:37:45.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_0/index_4516_seed_18835_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:38:09.462 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 00:38:10.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.611 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 337 train_steps 9480 {'step': '337/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00577091', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '294.88381958', 'grpo_loop_time': '421.08105469', 'ref_model_time': '110.92140961', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86469311', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23266602', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.383e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00918627', 'kl_moving': '0.00918627', 'kl_stepwise': '-1.918e-08', 'gathered_avg_reward_mean': '9.23010254', 'gathered_avg_reward_std': '1.46296227', 'gathered_video_score2_mq_reward_mean': '3.17419434', 'gathered_video_score2_mq_reward_std': '0.51533949', 'gathered_video_score2_overall_reward_mean': '9.23266602', 'gathered_video_score2_overall_reward_std': '1.46595037', 'gathered_video_score2_ta_reward_mean': '2.98559570', 'gathered_video_score2_ta_reward_std': '0.73211217', 'gathered_video_score2_vq_reward_mean': '3.07031250', 'gathered_video_score2_vq_reward_std': '0.52869296', 'gathered_ori_avg_reward_mean': '9.23010254', 'gathered_ori_avg_reward_std': '1.46296239'} +2026-06-28 00:49:02.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 00:49:02.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:09.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_0/index_1564_seed_28196_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 00:50:24.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:24.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:36.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_0/index_1564_seed_28197_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:51:56.392 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 00:51:57.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.284 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 338 train_steps 9520 {'step': '338/10000', 'learning_rate': '0.00001000', 'loss': '-4.470e-09', 'grad_norm': '0.00529773', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.13671875', 'grpo_loop_time': '419.67715454', 'ref_model_time': '112.84281921', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87771761', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.62988281', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.450e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.470e-09', 'kl_loss': '0.00928120', 'kl_moving': '0.00928120', 'kl_stepwise': '-1.915e-08', 'gathered_avg_reward_mean': '9.63269043', 'gathered_avg_reward_std': '1.08051109', 'gathered_video_score2_mq_reward_mean': '3.28808594', 'gathered_video_score2_mq_reward_std': '0.36661437', 'gathered_video_score2_overall_reward_mean': '9.62988281', 'gathered_video_score2_overall_reward_std': '1.07951415', 'gathered_video_score2_ta_reward_mean': '3.18750000', 'gathered_video_score2_ta_reward_std': '0.60392475', 'gathered_video_score2_vq_reward_mean': '3.15710449', 'gathered_video_score2_vq_reward_std': '0.45417637', 'gathered_ori_avg_reward_mean': '9.63269043', 'gathered_ori_avg_reward_std': '1.08051109'} +2026-06-28 01:02:48.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 01:02:48.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:54.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_0/index_7344_seed_23074_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 01:04:08.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:04:08.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:21.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_0/index_7344_seed_23075_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:05:43.979 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 01:05:44.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.190 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 339 train_steps 9560 {'step': '339/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00482241', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.36679077', 'grpo_loop_time': '418.26232910', 'ref_model_time': '111.12972260', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85233498', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.71142578', 'token_credit_video_advantage_mean': '-4.191e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.467e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.794e-09', 'kl_loss': '0.00964359', 'kl_moving': '0.00964359', 'kl_stepwise': '-1.992e-08', 'gathered_avg_reward_mean': '9.71325684', 'gathered_avg_reward_std': '1.20668411', 'gathered_video_score2_mq_reward_mean': '3.28051758', 'gathered_video_score2_mq_reward_std': '0.36560944', 'gathered_video_score2_overall_reward_mean': '9.71142578', 'gathered_video_score2_overall_reward_std': '1.20622504', 'gathered_video_score2_ta_reward_mean': '3.36938477', 'gathered_video_score2_ta_reward_std': '0.46436232', 'gathered_video_score2_vq_reward_mean': '3.06335449', 'gathered_video_score2_vq_reward_std': '0.59705299', 'gathered_ori_avg_reward_mean': '9.71325684', 'gathered_ori_avg_reward_std': '1.20668411'} +2026-06-28 01:16:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 01:16:33.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:39.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_0/index_1298_seed_19430_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 01:17:54.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:54.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:11.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_0/index_1298_seed_19431_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:19:30.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 01:19:31.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:21.283 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 340 train_steps 9600 {'step': '340/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00450227', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.04220581', 'grpo_loop_time': '419.56781006', 'ref_model_time': '111.06921387', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86309773', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01025391', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.457e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00961057', 'kl_moving': '0.00961057', 'kl_stepwise': '-1.953e-08', 'gathered_avg_reward_mean': '9.00646973', 'gathered_avg_reward_std': '1.10392833', 'gathered_video_score2_mq_reward_mean': '3.14526367', 'gathered_video_score2_mq_reward_std': '0.39109033', 'gathered_video_score2_overall_reward_mean': '9.01025391', 'gathered_video_score2_overall_reward_std': '1.10415053', 'gathered_video_score2_ta_reward_mean': '2.85656738', 'gathered_video_score2_ta_reward_std': '0.60106987', 'gathered_video_score2_vq_reward_mean': '3.00463867', 'gathered_video_score2_vq_reward_std': '0.49875242', 'gathered_ori_avg_reward_mean': '9.00646973', 'gathered_ori_avg_reward_std': '1.10392821'} +2026-06-28 01:30:56.494 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 01:30:58.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_0/index_265_seed_12442_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 01:32:20.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:20.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:30.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_0/index_265_seed_12443_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:33:51.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 01:33:52.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.500 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 341 train_steps 9640 {'step': '341/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00655047', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.21313477', 'grpo_loop_time': '420.18832397', 'ref_model_time': '111.26515961', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84407407', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.18798828', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.393e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00909473', 'kl_moving': '0.00909473', 'kl_stepwise': '-1.867e-08', 'gathered_avg_reward_mean': '9.18872070', 'gathered_avg_reward_std': '1.08518565', 'gathered_video_score2_mq_reward_mean': '3.18603516', 'gathered_video_score2_mq_reward_std': '0.32443893', 'gathered_video_score2_overall_reward_mean': '9.18798828', 'gathered_video_score2_overall_reward_std': '1.08493221', 'gathered_video_score2_ta_reward_mean': '3.08630371', 'gathered_video_score2_ta_reward_std': '0.54341370', 'gathered_video_score2_vq_reward_mean': '2.91638184', 'gathered_video_score2_vq_reward_std': '0.45157897', 'gathered_ori_avg_reward_mean': '9.18872070', 'gathered_ori_avg_reward_std': '1.08518577'} +2026-06-28 01:44:44.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 01:44:44.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:50.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_0/index_4381_seed_20764_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 01:46:04.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:46:04.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:16.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_0/index_4381_seed_20765_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:47:36.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 01:47:37.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.341 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 342 train_steps 9680 {'step': '342/10000', 'learning_rate': '0.00001000', 'loss': '-1.024e-08', 'grad_norm': '0.00590351', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.48684692', 'grpo_loop_time': '418.59204102', 'ref_model_time': '111.59663391', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88339806', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.50195312', 'token_credit_video_advantage_mean': '8.848e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.434e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.024e-08', 'kl_loss': '0.00833935', 'kl_moving': '0.00833935', 'kl_stepwise': '-1.899e-08', 'gathered_avg_reward_mean': '9.49963379', 'gathered_avg_reward_std': '1.09429955', 'gathered_video_score2_mq_reward_mean': '3.23925781', 'gathered_video_score2_mq_reward_std': '0.41846928', 'gathered_video_score2_overall_reward_mean': '9.50195312', 'gathered_video_score2_overall_reward_std': '1.09486377', 'gathered_video_score2_ta_reward_mean': '3.15502930', 'gathered_video_score2_ta_reward_std': '0.58003080', 'gathered_video_score2_vq_reward_mean': '3.10534668', 'gathered_video_score2_vq_reward_std': '0.42328322', 'gathered_ori_avg_reward_mean': '9.49963379', 'gathered_ori_avg_reward_std': '1.09429955'} +2026-06-28 01:58:27.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 01:58:27.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:33.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_0/index_8990_seed_29353_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 01:59:48.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:48.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:00:59.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_0/index_8990_seed_29354_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:01:17.857 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 02:01:18.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.548 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 343 train_steps 9720 {'step': '343/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00603537', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.63934326', 'grpo_loop_time': '419.72500610', 'ref_model_time': '111.31153870', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89515954', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.51367188', 'token_credit_video_advantage_mean': '-8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.471e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00865449', 'kl_moving': '0.00865449', 'kl_stepwise': '-1.929e-08', 'gathered_avg_reward_mean': '9.51367188', 'gathered_avg_reward_std': '0.83579832', 'gathered_video_score2_mq_reward_mean': '3.23095703', 'gathered_video_score2_mq_reward_std': '0.28305450', 'gathered_video_score2_overall_reward_mean': '9.51367188', 'gathered_video_score2_overall_reward_std': '0.84363925', 'gathered_video_score2_ta_reward_mean': '3.11816406', 'gathered_video_score2_ta_reward_std': '0.51386774', 'gathered_video_score2_vq_reward_mean': '3.16455078', 'gathered_video_score2_vq_reward_std': '0.34760103', 'gathered_ori_avg_reward_mean': '9.51367188', 'gathered_ori_avg_reward_std': '0.83579832'} +2026-06-28 02:12:08.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 02:12:08.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:14.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_0/index_4558_seed_13493_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 02:13:28.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:28.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:14:41.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_0/index_4558_seed_13494_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:15:02.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 02:15:04.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.569 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.230 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 344 train_steps 9760 {'step': '344/10000', 'learning_rate': '0.00001000', 'loss': '-5.122e-09', 'grad_norm': '0.00528150', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.22506714', 'grpo_loop_time': '419.59475708', 'ref_model_time': '110.99041748', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88311028', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.42822266', 'token_credit_video_advantage_mean': '4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.406e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.122e-09', 'kl_loss': '0.00890927', 'kl_moving': '0.00890927', 'kl_stepwise': '-1.923e-08', 'gathered_avg_reward_mean': '9.42785645', 'gathered_avg_reward_std': '0.76631194', 'gathered_video_score2_mq_reward_mean': '3.24658203', 'gathered_video_score2_mq_reward_std': '0.31125814', 'gathered_video_score2_overall_reward_mean': '9.42822266', 'gathered_video_score2_overall_reward_std': '0.76585776', 'gathered_video_score2_ta_reward_mean': '3.10498047', 'gathered_video_score2_ta_reward_std': '0.39116630', 'gathered_video_score2_vq_reward_mean': '3.07629395', 'gathered_video_score2_vq_reward_std': '0.38279441', 'gathered_ori_avg_reward_mean': '9.42785645', 'gathered_ori_avg_reward_std': '0.76631194'} +2026-06-28 02:25:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 02:25:53.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:59.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_0/index_7706_seed_3101_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 02:27:15.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:27:15.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:26.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_0/index_7706_seed_3102_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:28:44.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 02:28:45.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.484 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 345 train_steps 9800 {'step': '345/10000', 'learning_rate': '0.00001000', 'loss': '-1.490e-08', 'grad_norm': '0.00595135', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.47308350', 'grpo_loop_time': '420.27859497', 'ref_model_time': '111.32492828', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84349906', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.46679688', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.418e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.490e-08', 'kl_loss': '0.00918811', 'kl_moving': '0.00918811', 'kl_stepwise': '-1.977e-08', 'gathered_avg_reward_mean': '9.46911621', 'gathered_avg_reward_std': '0.98961532', 'gathered_video_score2_mq_reward_mean': '3.19042969', 'gathered_video_score2_mq_reward_std': '0.39548612', 'gathered_video_score2_overall_reward_mean': '9.46679688', 'gathered_video_score2_overall_reward_std': '0.99387580', 'gathered_video_score2_ta_reward_mean': '3.10388184', 'gathered_video_score2_ta_reward_std': '0.58488137', 'gathered_video_score2_vq_reward_mean': '3.17480469', 'gathered_video_score2_vq_reward_std': '0.44133604', 'gathered_ori_avg_reward_mean': '9.46911621', 'gathered_ori_avg_reward_std': '0.98961532'} +2026-06-28 02:39:38.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 02:39:38.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:44.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_0/index_8554_seed_16717_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 02:40:57.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:57.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:08.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_0/index_8554_seed_16718_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:42:30.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 02:42:31.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.676 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 346 train_steps 9840 {'step': '346/10000', 'learning_rate': '0.00001000', 'loss': '9.779e-09', 'grad_norm': '0.00537742', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.27481079', 'grpo_loop_time': '419.26785278', 'ref_model_time': '110.87035370', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83668572', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.40087891', 'token_credit_video_advantage_mean': '-3.492e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.431e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.779e-09', 'kl_loss': '0.00904954', 'kl_moving': '0.00904954', 'kl_stepwise': '-1.875e-08', 'gathered_avg_reward_mean': '9.40185547', 'gathered_avg_reward_std': '0.92088330', 'gathered_video_score2_mq_reward_mean': '3.20996094', 'gathered_video_score2_mq_reward_std': '0.38801742', 'gathered_video_score2_overall_reward_mean': '9.40087891', 'gathered_video_score2_overall_reward_std': '0.92261219', 'gathered_video_score2_ta_reward_mean': '3.10595703', 'gathered_video_score2_ta_reward_std': '0.47180375', 'gathered_video_score2_vq_reward_mean': '3.08593750', 'gathered_video_score2_vq_reward_std': '0.39711407', 'gathered_ori_avg_reward_mean': '9.40185547', 'gathered_ori_avg_reward_std': '0.92088330'} +2026-06-28 02:53:20.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 02:53:20.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_0/index_5793_seed_18299_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 02:54:40.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:40.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:55:52.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_0/index_5793_seed_18300_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:56:12.569 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 02:56:13.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.312 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:07.911 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 347 train_steps 9880 {'step': '347/10000', 'learning_rate': '0.00001000', 'loss': '1.304e-08', 'grad_norm': '0.00527872', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.57693481', 'grpo_loop_time': '420.67376709', 'ref_model_time': '113.08499146', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88954604', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23779297', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.405e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.304e-08', 'kl_loss': '0.01024406', 'kl_moving': '0.01024407', 'kl_stepwise': '-1.953e-08', 'gathered_avg_reward_mean': '9.23815918', 'gathered_avg_reward_std': '1.02341330', 'gathered_video_score2_mq_reward_mean': '3.19311523', 'gathered_video_score2_mq_reward_std': '0.35284454', 'gathered_video_score2_overall_reward_mean': '9.23779297', 'gathered_video_score2_overall_reward_std': '1.02480257', 'gathered_video_score2_ta_reward_mean': '2.92614746', 'gathered_video_score2_ta_reward_std': '0.67224956', 'gathered_video_score2_vq_reward_mean': '3.11889648', 'gathered_video_score2_vq_reward_std': '0.30313879', 'gathered_ori_avg_reward_mean': '9.23815918', 'gathered_ori_avg_reward_std': '1.02341330'} +2026-06-28 03:07:08.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 03:07:08.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:14.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_0/index_2705_seed_9578_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 03:08:31.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:31.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:09:38.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_0/index_2705_seed_9579_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:10:01.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 03:10:03.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.213 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 348 train_steps 9920 {'step': '348/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00574253', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.32000732', 'grpo_loop_time': '419.99310303', 'ref_model_time': '111.31849670', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87311351', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.29492188', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.363e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.01016704', 'kl_moving': '0.01016705', 'kl_stepwise': '-1.881e-08', 'gathered_avg_reward_mean': '9.29626465', 'gathered_avg_reward_std': '1.19701016', 'gathered_video_score2_mq_reward_mean': '3.25122070', 'gathered_video_score2_mq_reward_std': '0.42131677', 'gathered_video_score2_overall_reward_mean': '9.29492188', 'gathered_video_score2_overall_reward_std': '1.19617867', 'gathered_video_score2_ta_reward_mean': '2.94384766', 'gathered_video_score2_ta_reward_std': '0.63500875', 'gathered_video_score2_vq_reward_mean': '3.10119629', 'gathered_video_score2_vq_reward_std': '0.48188671', 'gathered_ori_avg_reward_mean': '9.29626465', 'gathered_ori_avg_reward_std': '1.19701016'} +2026-06-28 03:20:53.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 03:20:53.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:59.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_0/index_8757_seed_31171_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 03:22:15.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:22:15.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:26.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_0/index_8757_seed_31172_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:23:47.772 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 03:23:49.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.093 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 349 train_steps 9960 {'step': '349/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00765111', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.43966675', 'grpo_loop_time': '419.79818726', 'ref_model_time': '111.87602234', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86378634', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.32421875', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.427e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-10', 'kl_loss': '0.01072094', 'kl_moving': '0.01072094', 'kl_stepwise': '-1.921e-08', 'gathered_avg_reward_mean': '9.32666016', 'gathered_avg_reward_std': '1.03609085', 'gathered_video_score2_mq_reward_mean': '3.20166016', 'gathered_video_score2_mq_reward_std': '0.33064646', 'gathered_video_score2_overall_reward_mean': '9.32421875', 'gathered_video_score2_overall_reward_std': '1.04159260', 'gathered_video_score2_ta_reward_mean': '3.12451172', 'gathered_video_score2_ta_reward_std': '0.47177547', 'gathered_video_score2_vq_reward_mean': '3.00048828', 'gathered_video_score2_vq_reward_std': '0.46898568', 'gathered_ori_avg_reward_mean': '9.32666016', 'gathered_ori_avg_reward_std': '1.03609085'} +2026-06-28 03:34:40.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 03:34:40.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_0/index_6603_seed_20511_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 03:36:01.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:36:01.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:12.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_0/index_6603_seed_20512_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 03:37:34.480 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:37:34.651 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 03:37:35.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:48:26.399 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 350 train_steps 10000 {'step': '350/10000', 'learning_rate': '0.00001000', 'loss': '-2.235e-08', 'grad_norm': '0.00489396', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.48303223', 'grpo_loop_time': '419.15325928', 'ref_model_time': '113.31278992', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85318011', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.90917969', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.387e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.235e-08', 'kl_loss': '0.00994981', 'kl_moving': '0.00994982', 'kl_stepwise': '-1.884e-08', 'gathered_avg_reward_mean': '8.91003418', 'gathered_avg_reward_std': '1.25475323', 'gathered_video_score2_mq_reward_mean': '3.15478516', 'gathered_video_score2_mq_reward_std': '0.36037901', 'gathered_video_score2_overall_reward_mean': '8.90917969', 'gathered_video_score2_overall_reward_std': '1.25171661', 'gathered_video_score2_ta_reward_mean': '2.71911621', 'gathered_video_score2_ta_reward_std': '0.69493580', 'gathered_video_score2_vq_reward_mean': '3.03613281', 'gathered_video_score2_vq_reward_std': '0.42751470', 'gathered_ori_avg_reward_mean': '8.91003418', 'gathered_ori_avg_reward_std': '1.25475323'} +2026-06-28 03:49:33.218 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 350 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-350 +2026-06-28 03:49:33.353 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 350: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-350 +2026-06-28 03:49:36.140 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-325 +2026-06-28 03:50:11.874 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 03:50:13.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:20.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_0/index_5371_seed_27895_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 03:51:36.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:36.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:52:46.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_0/index_5371_seed_27896_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:53:05.873 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 03:53:06.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.732 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 351 train_steps 10040 {'step': '351/10000', 'learning_rate': '0.00001000', 'loss': '1.956e-08', 'grad_norm': '0.00597701', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.17077637', 'grpo_loop_time': '423.31060791', 'ref_model_time': '111.32084656', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86759883', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.48925781', 'token_credit_video_advantage_mean': '-7.683e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.455e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.956e-08', 'kl_loss': '0.00877910', 'kl_moving': '0.00877910', 'kl_stepwise': '-2.016e-08', 'gathered_avg_reward_mean': '9.49365234', 'gathered_avg_reward_std': '1.14409029', 'gathered_video_score2_mq_reward_mean': '3.28735352', 'gathered_video_score2_mq_reward_std': '0.38210052', 'gathered_video_score2_overall_reward_mean': '9.48925781', 'gathered_video_score2_overall_reward_std': '1.14713764', 'gathered_video_score2_ta_reward_mean': '3.21093750', 'gathered_video_score2_ta_reward_std': '0.58825064', 'gathered_video_score2_vq_reward_mean': '2.99536133', 'gathered_video_score2_vq_reward_std': '0.50868076', 'gathered_ori_avg_reward_mean': '9.49365234', 'gathered_ori_avg_reward_std': '1.14409029'} +2026-06-28 04:04:02.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 04:04:02.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:09.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_0/index_4187_seed_25775_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 04:05:24.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:24.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:33.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_0/index_4187_seed_25776_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 04:06:55.563 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:06:55.729 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 04:06:56.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.511 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 352 train_steps 10080 {'step': '352/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00682441', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.10119629', 'grpo_loop_time': '419.50775146', 'ref_model_time': '113.61826324', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87091106', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.09375000', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00907637', 'kl_moving': '0.00907637', 'kl_stepwise': '-1.882e-08', 'gathered_avg_reward_mean': '9.09265137', 'gathered_avg_reward_std': '0.91119075', 'gathered_video_score2_mq_reward_mean': '3.11279297', 'gathered_video_score2_mq_reward_std': '0.33406395', 'gathered_video_score2_overall_reward_mean': '9.09375000', 'gathered_video_score2_overall_reward_std': '0.90831864', 'gathered_video_score2_ta_reward_mean': '3.02502441', 'gathered_video_score2_ta_reward_std': '0.65881163', 'gathered_video_score2_vq_reward_mean': '2.95483398', 'gathered_video_score2_vq_reward_std': '0.35568458', 'gathered_ori_avg_reward_mean': '9.09265137', 'gathered_ori_avg_reward_std': '0.91119075'} +2026-06-28 04:17:48.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 04:17:48.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_0/index_3906_seed_16283_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 04:19:09.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:19:09.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:20.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_0/index_3906_seed_16284_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:20:45.852 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 04:20:47.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.119 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 353 train_steps 10120 {'step': '353/10000', 'learning_rate': '0.00001000', 'loss': '-1.537e-08', 'grad_norm': '0.00615320', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.61764526', 'grpo_loop_time': '421.05627441', 'ref_model_time': '111.44221497', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87922525', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.62451172', 'token_credit_video_advantage_mean': '1.630e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.462e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.909e-08', 'kl_loss': '0.00948206', 'kl_moving': '0.00948206', 'kl_stepwise': '-1.935e-08', 'gathered_avg_reward_mean': '9.62487793', 'gathered_avg_reward_std': '0.88292539', 'gathered_video_score2_mq_reward_mean': '3.28833008', 'gathered_video_score2_mq_reward_std': '0.30276498', 'gathered_video_score2_overall_reward_mean': '9.62451172', 'gathered_video_score2_overall_reward_std': '0.88603002', 'gathered_video_score2_ta_reward_mean': '3.20349121', 'gathered_video_score2_ta_reward_std': '0.52505368', 'gathered_video_score2_vq_reward_mean': '3.13305664', 'gathered_video_score2_vq_reward_std': '0.34814957', 'gathered_ori_avg_reward_mean': '9.62487793', 'gathered_ori_avg_reward_std': '0.88292539'} +2026-06-28 04:31:40.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 04:31:40.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:46.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_0/index_2366_seed_26208_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 04:33:02.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:33:02.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:09.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_0/index_2366_seed_26209_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:34:28.746 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 04:34:29.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.259 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 354 train_steps 10160 {'step': '354/10000', 'learning_rate': '0.00001000', 'loss': '2.608e-08', 'grad_norm': '0.00522298', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '281.63095093', 'grpo_loop_time': '419.20004272', 'ref_model_time': '111.45374298', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83793068', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.74462891', 'token_credit_video_advantage_mean': '-8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.434e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-08', 'kl_loss': '0.00875082', 'kl_moving': '0.00875082', 'kl_stepwise': '-1.936e-08', 'gathered_avg_reward_mean': '8.74487305', 'gathered_avg_reward_std': '0.99317622', 'gathered_video_score2_mq_reward_mean': '3.06835938', 'gathered_video_score2_mq_reward_std': '0.28565747', 'gathered_video_score2_overall_reward_mean': '8.74462891', 'gathered_video_score2_overall_reward_std': '0.99345630', 'gathered_video_score2_ta_reward_mean': '2.71875000', 'gathered_video_score2_ta_reward_std': '0.59109867', 'gathered_video_score2_vq_reward_mean': '2.95776367', 'gathered_video_score2_vq_reward_std': '0.41541165', 'gathered_ori_avg_reward_mean': '8.74487305', 'gathered_ori_avg_reward_std': '0.99317628'} +2026-06-28 04:45:20.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 04:45:20.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:26.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_0/index_2069_seed_16967_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 04:46:45.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:45.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:47:51.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_0/index_2069_seed_16968_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:48:10.183 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 04:48:11.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.039 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 355 train_steps 10200 {'step': '355/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-10', 'grad_norm': '0.00453324', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.49905396', 'grpo_loop_time': '421.62115479', 'ref_model_time': '111.98599243', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86344337', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23144531', 'token_credit_video_advantage_mean': '6.985e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.433e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.985e-10', 'kl_loss': '0.00874050', 'kl_moving': '0.00874050', 'kl_stepwise': '-1.951e-08', 'gathered_avg_reward_mean': '9.23083496', 'gathered_avg_reward_std': '1.11771739', 'gathered_video_score2_mq_reward_mean': '3.13916016', 'gathered_video_score2_mq_reward_std': '0.39007500', 'gathered_video_score2_overall_reward_mean': '9.23144531', 'gathered_video_score2_overall_reward_std': '1.12106955', 'gathered_video_score2_ta_reward_mean': '3.03137207', 'gathered_video_score2_ta_reward_std': '0.62322515', 'gathered_video_score2_vq_reward_mean': '3.06030273', 'gathered_video_score2_vq_reward_std': '0.48589745', 'gathered_ori_avg_reward_mean': '9.23083496', 'gathered_ori_avg_reward_std': '1.11771739'} +2026-06-28 04:59:03.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 04:59:03.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:09.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_0/index_9974_seed_5757_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 05:00:24.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:24.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:32.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_0/index_9974_seed_5758_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:01:51.600 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 05:01:52.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.663 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 356 train_steps 10240 {'step': '356/10000', 'learning_rate': '0.00001000', 'loss': '2.328e-09', 'grad_norm': '0.00652245', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '281.54882812', 'grpo_loop_time': '420.11013794', 'ref_model_time': '111.42153168', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85311621', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.47558594', 'token_credit_video_advantage_mean': '-2.328e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.454e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00991854', 'kl_moving': '0.00991854', 'kl_stepwise': '-2.066e-08', 'gathered_avg_reward_mean': '9.47485352', 'gathered_avg_reward_std': '0.95288229', 'gathered_video_score2_mq_reward_mean': '3.30004883', 'gathered_video_score2_mq_reward_std': '0.33722025', 'gathered_video_score2_overall_reward_mean': '9.47558594', 'gathered_video_score2_overall_reward_std': '0.95450830', 'gathered_video_score2_ta_reward_mean': '2.94433594', 'gathered_video_score2_ta_reward_std': '0.58408314', 'gathered_video_score2_vq_reward_mean': '3.23046875', 'gathered_video_score2_vq_reward_std': '0.42970526', 'gathered_ori_avg_reward_mean': '9.47485352', 'gathered_ori_avg_reward_std': '0.95288229'} +2026-06-28 05:12:44.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 05:12:44.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_0/index_5936_seed_30179_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 05:14:03.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:14:03.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:15.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_0/index_5936_seed_30180_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:15:34.182 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 05:15:34.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.747 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 357 train_steps 10280 {'step': '357/10000', 'learning_rate': '0.00001000', 'loss': '1.094e-08', 'grad_norm': '0.00735178', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.71530151', 'grpo_loop_time': '420.21801758', 'ref_model_time': '111.48618317', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85896504', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.13720703', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.421e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.094e-08', 'kl_loss': '0.01056590', 'kl_moving': '0.01056590', 'kl_stepwise': '-1.927e-08', 'gathered_avg_reward_mean': '9.13146973', 'gathered_avg_reward_std': '1.24628854', 'gathered_video_score2_mq_reward_mean': '3.19921875', 'gathered_video_score2_mq_reward_std': '0.44112095', 'gathered_video_score2_overall_reward_mean': '9.13720703', 'gathered_video_score2_overall_reward_std': '1.24403632', 'gathered_video_score2_ta_reward_mean': '2.88598633', 'gathered_video_score2_ta_reward_std': '0.68231845', 'gathered_video_score2_vq_reward_mean': '3.04626465', 'gathered_video_score2_vq_reward_std': '0.51718795', 'gathered_ori_avg_reward_mean': '9.13146973', 'gathered_ori_avg_reward_std': '1.24628854'} +2026-06-28 05:26:25.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 05:26:25.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:32.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_0/index_8817_seed_8397_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 05:27:47.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:47.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:28:57.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_0/index_8817_seed_8398_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:29:21.209 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 05:29:22.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.874 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 358 train_steps 10320 {'step': '358/10000', 'learning_rate': '0.00001000', 'loss': '5.122e-09', 'grad_norm': '0.00548495', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.44604492', 'grpo_loop_time': '420.44329834', 'ref_model_time': '111.35923767', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86660266', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.16162109', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.478e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.122e-09', 'kl_loss': '0.01079848', 'kl_moving': '0.01079848', 'kl_stepwise': '-1.946e-08', 'gathered_avg_reward_mean': '9.16247559', 'gathered_avg_reward_std': '1.00111341', 'gathered_video_score2_mq_reward_mean': '3.19946289', 'gathered_video_score2_mq_reward_std': '0.36686262', 'gathered_video_score2_overall_reward_mean': '9.16162109', 'gathered_video_score2_overall_reward_std': '1.00600243', 'gathered_video_score2_ta_reward_mean': '2.97253418', 'gathered_video_score2_ta_reward_std': '0.61234593', 'gathered_video_score2_vq_reward_mean': '2.99047852', 'gathered_video_score2_vq_reward_std': '0.41093320', 'gathered_ori_avg_reward_mean': '9.16247559', 'gathered_ori_avg_reward_std': '1.00111341'} +2026-06-28 05:40:14.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 05:40:15.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_0/index_8191_seed_16895_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 05:41:35.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:35.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:42:47.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_0/index_8191_seed_16896_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:43:10.017 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 05:43:10.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.698 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 359 train_steps 10360 {'step': '359/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00455104', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.81930542', 'grpo_loop_time': '420.47698975', 'ref_model_time': '114.09541321', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84970361', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.30810547', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.396e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.01064670', 'kl_moving': '0.01064670', 'kl_stepwise': '-1.916e-08', 'gathered_avg_reward_mean': '9.30957031', 'gathered_avg_reward_std': '0.97528774', 'gathered_video_score2_mq_reward_mean': '3.17272949', 'gathered_video_score2_mq_reward_std': '0.35670105', 'gathered_video_score2_overall_reward_mean': '9.30810547', 'gathered_video_score2_overall_reward_std': '0.97523141', 'gathered_video_score2_ta_reward_mean': '3.09948730', 'gathered_video_score2_ta_reward_std': '0.52281266', 'gathered_video_score2_vq_reward_mean': '3.03735352', 'gathered_video_score2_vq_reward_std': '0.35728437', 'gathered_ori_avg_reward_mean': '9.30957031', 'gathered_ori_avg_reward_std': '0.97528774'} +2026-06-28 05:54:06.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 05:54:06.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:13.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_0/index_509_seed_2399_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 05:55:33.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:33.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:38.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_0/index_509_seed_2400_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:56:59.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 05:57:00.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:07:49.401 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 360 train_steps 10400 {'step': '360/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-10', 'grad_norm': '0.00548784', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.76718140', 'grpo_loop_time': '418.10205078', 'ref_model_time': '112.24012756', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86114156', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.07324219', 'token_credit_video_advantage_mean': '1.187e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.450e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-10', 'kl_loss': '0.01035270', 'kl_moving': '0.01035270', 'kl_stepwise': '-1.913e-08', 'gathered_avg_reward_mean': '9.07727051', 'gathered_avg_reward_std': '1.00353491', 'gathered_video_score2_mq_reward_mean': '3.13964844', 'gathered_video_score2_mq_reward_std': '0.28536853', 'gathered_video_score2_overall_reward_mean': '9.07324219', 'gathered_video_score2_overall_reward_std': '1.00396276', 'gathered_video_score2_ta_reward_mean': '2.85803223', 'gathered_video_score2_ta_reward_std': '0.61518252', 'gathered_video_score2_vq_reward_mean': '3.07958984', 'gathered_video_score2_vq_reward_std': '0.41308218', 'gathered_ori_avg_reward_mean': '9.07727051', 'gathered_ori_avg_reward_std': '1.00353491'} +2026-06-28 06:08:24.976 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 06:08:27.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:33.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_0/index_9998_seed_26043_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 06:09:47.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:47.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:10:59.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_0/index_9998_seed_26044_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 06:11:20.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:11:20.368 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 06:11:21.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.579 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 361 train_steps 10440 {'step': '361/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00548951', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.40225220', 'grpo_loop_time': '419.35183716', 'ref_model_time': '111.47416687', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88065445', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.58496094', 'token_credit_video_advantage_mean': '2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.480e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00824581', 'kl_moving': '0.00824581', 'kl_stepwise': '-1.916e-08', 'gathered_avg_reward_mean': '9.58740234', 'gathered_avg_reward_std': '1.19534993', 'gathered_video_score2_mq_reward_mean': '3.27416992', 'gathered_video_score2_mq_reward_std': '0.53541052', 'gathered_video_score2_overall_reward_mean': '9.58496094', 'gathered_video_score2_overall_reward_std': '1.19245565', 'gathered_video_score2_ta_reward_mean': '3.17163086', 'gathered_video_score2_ta_reward_std': '0.67894763', 'gathered_video_score2_vq_reward_mean': '3.14160156', 'gathered_video_score2_vq_reward_std': '0.42396650', 'gathered_ori_avg_reward_mean': '9.58740234', 'gathered_ori_avg_reward_std': '1.19534993'} +2026-06-28 06:22:11.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 06:22:11.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:18.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_0/index_5544_seed_23712_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 06:23:32.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:32.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:24:42.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_0/index_5544_seed_23713_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:25:06.234 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 06:25:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:01.918 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 362 train_steps 10480 {'step': '362/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00626851', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.41589355', 'grpo_loop_time': '422.62652588', 'ref_model_time': '111.19480133', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89982074', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.17968750', 'token_credit_video_advantage_mean': '8.848e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.487e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.00906999', 'kl_moving': '0.00906999', 'kl_stepwise': '-2.059e-08', 'gathered_avg_reward_mean': '9.17639160', 'gathered_avg_reward_std': '1.20922244', 'gathered_video_score2_mq_reward_mean': '3.18994141', 'gathered_video_score2_mq_reward_std': '0.37437099', 'gathered_video_score2_overall_reward_mean': '9.17968750', 'gathered_video_score2_overall_reward_std': '1.21213400', 'gathered_video_score2_ta_reward_mean': '2.88195801', 'gathered_video_score2_ta_reward_std': '0.63791007', 'gathered_video_score2_vq_reward_mean': '3.10449219', 'gathered_video_score2_vq_reward_std': '0.58390683', 'gathered_ori_avg_reward_mean': '9.17639160', 'gathered_ori_avg_reward_std': '1.20922244'} +2026-06-28 06:36:02.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 06:36:02.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:08.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_0/index_2488_seed_16835_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 06:37:22.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:22.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:32.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_0/index_2488_seed_16836_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:38:52.719 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 06:38:53.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.450 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 363 train_steps 10520 {'step': '363/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00652951', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.76489258', 'grpo_loop_time': '420.79421997', 'ref_model_time': '111.45957947', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88752383', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.42089844', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.408e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00847088', 'kl_moving': '0.00847088', 'kl_stepwise': '-1.978e-08', 'gathered_avg_reward_mean': '9.41931152', 'gathered_avg_reward_std': '1.41242671', 'gathered_video_score2_mq_reward_mean': '3.21362305', 'gathered_video_score2_mq_reward_std': '0.42545012', 'gathered_video_score2_overall_reward_mean': '9.42089844', 'gathered_video_score2_overall_reward_std': '1.40903556', 'gathered_video_score2_ta_reward_mean': '3.11926270', 'gathered_video_score2_ta_reward_std': '0.61562061', 'gathered_video_score2_vq_reward_mean': '3.08642578', 'gathered_video_score2_vq_reward_std': '0.51410031', 'gathered_ori_avg_reward_mean': '9.41931152', 'gathered_ori_avg_reward_std': '1.41242683'} +2026-06-28 06:49:45.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 06:49:45.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:51.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_0/index_1686_seed_18231_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 06:51:05.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:51:05.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:16.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_0/index_1686_seed_18232_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 06:52:36.807 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:52:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 06:52:37.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.606 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 364 train_steps 10560 {'step': '364/10000', 'learning_rate': '0.00001000', 'loss': '1.770e-08', 'grad_norm': '0.00789288', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.54821777', 'grpo_loop_time': '419.75299072', 'ref_model_time': '111.50379944', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88022661', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.88964844', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.770e-08', 'kl_loss': '0.00911073', 'kl_moving': '0.00911074', 'kl_stepwise': '-1.923e-08', 'gathered_avg_reward_mean': '8.88647461', 'gathered_avg_reward_std': '1.00053859', 'gathered_video_score2_mq_reward_mean': '3.08593750', 'gathered_video_score2_mq_reward_std': '0.40451854', 'gathered_video_score2_overall_reward_mean': '8.88964844', 'gathered_video_score2_overall_reward_std': '1.00103652', 'gathered_video_score2_ta_reward_mean': '2.66943359', 'gathered_video_score2_ta_reward_std': '0.66773868', 'gathered_video_score2_vq_reward_mean': '3.13110352', 'gathered_video_score2_vq_reward_std': '0.41749954', 'gathered_ori_avg_reward_mean': '8.88647461', 'gathered_ori_avg_reward_std': '1.00053871'} +2026-06-28 07:03:27.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 07:03:27.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:33.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_0/index_6991_seed_16871_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 07:04:51.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:51.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:00.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_0/index_6991_seed_16872_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:06:20.786 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 07:06:21.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.010 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 365 train_steps 10600 {'step': '365/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00500919', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.21783447', 'grpo_loop_time': '419.18191528', 'ref_model_time': '111.45523071', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85556304', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.51757812', 'token_credit_video_advantage_mean': '6.985e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.375e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00923982', 'kl_moving': '0.00923982', 'kl_stepwise': '-1.836e-08', 'gathered_avg_reward_mean': '9.51892090', 'gathered_avg_reward_std': '0.94369781', 'gathered_video_score2_mq_reward_mean': '3.25878906', 'gathered_video_score2_mq_reward_std': '0.36507115', 'gathered_video_score2_overall_reward_mean': '9.51757812', 'gathered_video_score2_overall_reward_std': '0.94293469', 'gathered_video_score2_ta_reward_mean': '3.16394043', 'gathered_video_score2_ta_reward_std': '0.54070795', 'gathered_video_score2_vq_reward_mean': '3.09619141', 'gathered_video_score2_vq_reward_std': '0.36324155', 'gathered_ori_avg_reward_mean': '9.51892090', 'gathered_ori_avg_reward_std': '0.94369781'} +2026-06-28 07:17:12.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 07:17:12.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:18.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_0/index_5531_seed_12769_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 07:18:35.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:35.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:19:43.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_0/index_5531_seed_12770_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:20:03.261 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 07:20:04.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.634 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 366 train_steps 10640 {'step': '366/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00450327', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.32791138', 'grpo_loop_time': '419.33737183', 'ref_model_time': '111.27842712', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88339609', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.52246094', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.431e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.00862817', 'kl_moving': '0.00862817', 'kl_stepwise': '-1.979e-08', 'gathered_avg_reward_mean': '9.51647949', 'gathered_avg_reward_std': '0.94473541', 'gathered_video_score2_mq_reward_mean': '3.31054688', 'gathered_video_score2_mq_reward_std': '0.37277067', 'gathered_video_score2_overall_reward_mean': '9.52246094', 'gathered_video_score2_overall_reward_std': '0.94968432', 'gathered_video_score2_ta_reward_mean': '2.96325684', 'gathered_video_score2_ta_reward_std': '0.52488703', 'gathered_video_score2_vq_reward_mean': '3.24267578', 'gathered_video_score2_vq_reward_std': '0.44824937', 'gathered_ori_avg_reward_mean': '9.51647949', 'gathered_ori_avg_reward_std': '0.94473541'} +2026-06-28 07:30:53.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 07:30:53.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:59.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_0/index_8773_seed_18105_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 07:32:18.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:32:18.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:26.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_0/index_8773_seed_18106_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:33:47.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 07:33:48.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.137 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 367 train_steps 10680 {'step': '367/10000', 'learning_rate': '0.00001000', 'loss': '-7.916e-09', 'grad_norm': '0.00704966', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.52255249', 'grpo_loop_time': '420.74703979', 'ref_model_time': '111.21742249', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84102124', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.56738281', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.492e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.916e-09', 'kl_loss': '0.01027057', 'kl_moving': '0.01027058', 'kl_stepwise': '-1.925e-08', 'gathered_avg_reward_mean': '9.56567383', 'gathered_avg_reward_std': '0.86376137', 'gathered_video_score2_mq_reward_mean': '3.24023438', 'gathered_video_score2_mq_reward_std': '0.42104003', 'gathered_video_score2_overall_reward_mean': '9.56738281', 'gathered_video_score2_overall_reward_std': '0.86138290', 'gathered_video_score2_ta_reward_mean': '3.20080566', 'gathered_video_score2_ta_reward_std': '0.52491790', 'gathered_video_score2_vq_reward_mean': '3.12463379', 'gathered_video_score2_vq_reward_std': '0.40108654', 'gathered_ori_avg_reward_mean': '9.56567383', 'gathered_ori_avg_reward_std': '0.86376137'} +2026-06-28 07:44:40.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 07:44:40.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:46.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_0/index_7853_seed_25652_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 07:46:01.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:46:02.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:12.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_0/index_7853_seed_25653_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:47:32.213 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 07:47:33.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.600 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 368 train_steps 10720 {'step': '368/10000', 'learning_rate': '0.00001000', 'loss': '-1.024e-08', 'grad_norm': '0.00580870', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.11044312', 'grpo_loop_time': '417.91851807', 'ref_model_time': '111.44688416', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86299580', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.98974609', 'token_credit_video_advantage_mean': '-6.985e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.423e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.024e-08', 'kl_loss': '0.00953882', 'kl_moving': '0.00953882', 'kl_stepwise': '-1.911e-08', 'gathered_avg_reward_mean': '8.98962402', 'gathered_avg_reward_std': '0.97384101', 'gathered_video_score2_mq_reward_mean': '3.13610840', 'gathered_video_score2_mq_reward_std': '0.39252907', 'gathered_video_score2_overall_reward_mean': '8.98974609', 'gathered_video_score2_overall_reward_std': '0.97316682', 'gathered_video_score2_ta_reward_mean': '2.86816406', 'gathered_video_score2_ta_reward_std': '0.54027390', 'gathered_video_score2_vq_reward_mean': '2.98535156', 'gathered_video_score2_vq_reward_std': '0.51774114', 'gathered_ori_avg_reward_mean': '8.98962402', 'gathered_ori_avg_reward_std': '0.97384101'} +2026-06-28 07:58:21.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 07:58:21.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_0/index_7626_seed_11731_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 07:59:45.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:45.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:00:53.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_0/index_7626_seed_11732_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:01:15.773 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 08:01:16.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.513 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 369 train_steps 10760 {'step': '369/10000', 'learning_rate': '0.00001000', 'loss': '2.794e-09', 'grad_norm': '0.00553495', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.55941772', 'grpo_loop_time': '420.37228394', 'ref_model_time': '112.73086548', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85097951', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '10.01220703', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.464e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.794e-09', 'kl_loss': '0.01016192', 'kl_moving': '0.01016192', 'kl_stepwise': '-2.044e-08', 'gathered_avg_reward_mean': '10.01269531', 'gathered_avg_reward_std': '1.08129728', 'gathered_video_score2_mq_reward_mean': '3.41284180', 'gathered_video_score2_mq_reward_std': '0.39143279', 'gathered_video_score2_overall_reward_mean': '10.01220703', 'gathered_video_score2_overall_reward_std': '1.07879877', 'gathered_video_score2_ta_reward_mean': '3.39672852', 'gathered_video_score2_ta_reward_std': '0.55581844', 'gathered_video_score2_vq_reward_mean': '3.20312500', 'gathered_video_score2_vq_reward_std': '0.46141726', 'gathered_ori_avg_reward_mean': '10.01269531', 'gathered_ori_avg_reward_std': '1.08129728'} +2026-06-28 08:12:08.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 08:12:08.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:15.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_0/index_481_seed_9610_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 08:13:28.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:28.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:14:39.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_0/index_481_seed_9611_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:15:02.421 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 08:15:03.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.523 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:25:54.112 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 370 train_steps 10800 {'step': '370/10000', 'learning_rate': '0.00001000', 'loss': '2.049e-08', 'grad_norm': '0.00646614', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.24591064', 'grpo_loop_time': '420.04025269', 'ref_model_time': '111.53286743', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85213608', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.22949219', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.049e-08', 'kl_loss': '0.01131383', 'kl_moving': '0.01131383', 'kl_stepwise': '-1.891e-08', 'gathered_avg_reward_mean': '9.22399902', 'gathered_avg_reward_std': '1.01698577', 'gathered_video_score2_mq_reward_mean': '3.20800781', 'gathered_video_score2_mq_reward_std': '0.32959673', 'gathered_video_score2_overall_reward_mean': '9.22949219', 'gathered_video_score2_overall_reward_std': '1.01951206', 'gathered_video_score2_ta_reward_mean': '3.05627441', 'gathered_video_score2_ta_reward_std': '0.49830756', 'gathered_video_score2_vq_reward_mean': '2.95971680', 'gathered_video_score2_vq_reward_std': '0.45169836', 'gathered_ori_avg_reward_mean': '9.22399902', 'gathered_ori_avg_reward_std': '1.01698565'} +2026-06-28 08:26:29.645 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 08:26:31.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:37.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_0/index_1385_seed_549_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 08:27:51.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:51.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:02.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_0/index_1385_seed_550_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:29:27.136 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 08:29:28.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.489 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 371 train_steps 10840 {'step': '371/10000', 'learning_rate': '0.00001000', 'loss': '2.049e-08', 'grad_norm': '0.00816296', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.81619263', 'grpo_loop_time': '422.43121338', 'ref_model_time': '111.39090729', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86522853', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.99169922', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.459e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.956e-08', 'kl_loss': '0.00904444', 'kl_moving': '0.00904444', 'kl_stepwise': '-2.016e-08', 'gathered_avg_reward_mean': '8.99206543', 'gathered_avg_reward_std': '1.55071461', 'gathered_video_score2_mq_reward_mean': '3.10876465', 'gathered_video_score2_mq_reward_std': '0.55140477', 'gathered_video_score2_overall_reward_mean': '8.99169922', 'gathered_video_score2_overall_reward_std': '1.55081809', 'gathered_video_score2_ta_reward_mean': '2.95947266', 'gathered_video_score2_ta_reward_std': '0.67870301', 'gathered_video_score2_vq_reward_mean': '2.92382812', 'gathered_video_score2_vq_reward_std': '0.56824023', 'gathered_ori_avg_reward_mean': '8.99206543', 'gathered_ori_avg_reward_std': '1.55071449'} +2026-06-28 08:40:22.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 08:40:22.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:29.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_0/index_2027_seed_2605_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 08:41:47.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:47.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:42:53.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_0/index_2027_seed_2606_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:43:13.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 08:43:14.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.700 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 372 train_steps 10880 {'step': '372/10000', 'learning_rate': '0.00001000', 'loss': '8.848e-09', 'grad_norm': '0.00536061', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.58248901', 'grpo_loop_time': '420.33050537', 'ref_model_time': '113.43048096', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86855292', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.12207031', 'token_credit_video_advantage_mean': '3.958e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.454e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.848e-09', 'kl_loss': '0.00868658', 'kl_moving': '0.00868658', 'kl_stepwise': '-1.905e-08', 'gathered_avg_reward_mean': '9.12231445', 'gathered_avg_reward_std': '1.15288007', 'gathered_video_score2_mq_reward_mean': '3.14331055', 'gathered_video_score2_mq_reward_std': '0.34072438', 'gathered_video_score2_overall_reward_mean': '9.12207031', 'gathered_video_score2_overall_reward_std': '1.15329969', 'gathered_video_score2_ta_reward_mean': '3.05090332', 'gathered_video_score2_ta_reward_std': '0.71140683', 'gathered_video_score2_vq_reward_mean': '2.92810059', 'gathered_video_score2_vq_reward_std': '0.50771517', 'gathered_ori_avg_reward_mean': '9.12231445', 'gathered_ori_avg_reward_std': '1.15288007'} +2026-06-28 08:54:07.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 08:54:07.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:07.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_0/index_8982_seed_2909_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 08:55:27.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:27.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:39.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_0/index_8982_seed_2910_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 08:56:59.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:56:59.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 08:57:00.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.653 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 373 train_steps 10920 {'step': '373/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00875634', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.59808350', 'grpo_loop_time': '419.35754395', 'ref_model_time': '112.41935730', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86925501', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.29492188', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.419e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00913898', 'kl_moving': '0.00913898', 'kl_stepwise': '-1.880e-08', 'gathered_avg_reward_mean': '9.29589844', 'gathered_avg_reward_std': '0.97492296', 'gathered_video_score2_mq_reward_mean': '3.17919922', 'gathered_video_score2_mq_reward_std': '0.30699459', 'gathered_video_score2_overall_reward_mean': '9.29492188', 'gathered_video_score2_overall_reward_std': '0.97587281', 'gathered_video_score2_ta_reward_mean': '3.08203125', 'gathered_video_score2_ta_reward_std': '0.52441859', 'gathered_video_score2_vq_reward_mean': '3.03466797', 'gathered_video_score2_vq_reward_std': '0.40845507', 'gathered_ori_avg_reward_mean': '9.29589844', 'gathered_ori_avg_reward_std': '0.97492296'} +2026-06-28 09:07:50.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 09:07:50.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_0/index_1228_seed_2425_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 09:09:09.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:09:09.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:21.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_0/index_1228_seed_2426_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 09:10:42.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:10:43.119 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 09:10:44.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.421 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 374 train_steps 10960 {'step': '374/10000', 'learning_rate': '0.00001000', 'loss': '-7.916e-09', 'grad_norm': '0.01031396', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.56204224', 'grpo_loop_time': '418.15402222', 'ref_model_time': '111.56217194', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.83771622', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.98535156', 'token_credit_video_advantage_mean': '8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.402e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.916e-09', 'kl_loss': '0.00965948', 'kl_moving': '0.00965949', 'kl_stepwise': '-1.774e-08', 'gathered_avg_reward_mean': '8.98925781', 'gathered_avg_reward_std': '0.84057528', 'gathered_video_score2_mq_reward_mean': '3.14477539', 'gathered_video_score2_mq_reward_std': '0.27769074', 'gathered_video_score2_overall_reward_mean': '8.98535156', 'gathered_video_score2_overall_reward_std': '0.84103239', 'gathered_video_score2_ta_reward_mean': '2.99963379', 'gathered_video_score2_ta_reward_std': '0.51643527', 'gathered_video_score2_vq_reward_mean': '2.84484863', 'gathered_video_score2_vq_reward_std': '0.41285354', 'gathered_ori_avg_reward_mean': '8.98925781', 'gathered_ori_avg_reward_std': '0.84057528'} +2026-06-28 09:21:32.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 09:21:32.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:38.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_0/index_467_seed_12323_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 09:22:53.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:53.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:05.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_0/index_467_seed_12324_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:24:27.699 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 09:24:28.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:35:20.302 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 375 train_steps 11000 {'step': '375/10000', 'learning_rate': '0.00001000', 'loss': '-2.794e-09', 'grad_norm': '0.00509096', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.65594482', 'grpo_loop_time': '419.66726685', 'ref_model_time': '111.84211731', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.90129948', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.73730469', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.515e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.794e-09', 'kl_loss': '0.00946623', 'kl_moving': '0.00946623', 'kl_stepwise': '-1.885e-08', 'gathered_avg_reward_mean': '9.73498535', 'gathered_avg_reward_std': '1.14770591', 'gathered_video_score2_mq_reward_mean': '3.41162109', 'gathered_video_score2_mq_reward_std': '0.39021677', 'gathered_video_score2_overall_reward_mean': '9.73730469', 'gathered_video_score2_overall_reward_std': '1.14804840', 'gathered_video_score2_ta_reward_mean': '3.09631348', 'gathered_video_score2_ta_reward_std': '0.68234271', 'gathered_video_score2_vq_reward_mean': '3.22705078', 'gathered_video_score2_vq_reward_std': '0.35572588', 'gathered_ori_avg_reward_mean': '9.73498535', 'gathered_ori_avg_reward_std': '1.14770591'} +2026-06-28 09:36:32.001 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 375 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-375 +2026-06-28 09:36:32.148 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 375: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-375 +2026-06-28 09:36:34.976 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-350 +2026-06-28 09:36:34.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 09:36:35.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:41.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_0/index_1779_seed_23101_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 09:37:58.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:58.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:05.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_0/index_1779_seed_23102_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 09:39:26.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:39:26.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 09:39:27.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.360 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 376 train_steps 11040 {'step': '376/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00628392', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.45333862', 'grpo_loop_time': '422.29626465', 'ref_model_time': '111.89017487', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86184645', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.34765625', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.437e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.01001266', 'kl_moving': '0.01001267', 'kl_stepwise': '-1.925e-08', 'gathered_avg_reward_mean': '9.34777832', 'gathered_avg_reward_std': '0.93248361', 'gathered_video_score2_mq_reward_mean': '3.22998047', 'gathered_video_score2_mq_reward_std': '0.30392870', 'gathered_video_score2_overall_reward_mean': '9.34765625', 'gathered_video_score2_overall_reward_std': '0.93395323', 'gathered_video_score2_ta_reward_mean': '3.02795410', 'gathered_video_score2_ta_reward_std': '0.44452113', 'gathered_video_score2_vq_reward_mean': '3.08984375', 'gathered_video_score2_vq_reward_std': '0.45068544', 'gathered_ori_avg_reward_mean': '9.34777832', 'gathered_ori_avg_reward_std': '0.93248373'} +2026-06-28 09:50:22.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 09:50:22.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:28.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_0/index_6722_seed_22497_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 09:51:43.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:43.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:52:55.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_0/index_6722_seed_22498_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:53:24.118 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 09:53:25.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.615 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.283 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 377 train_steps 11080 {'step': '377/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00592010', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '295.37255859', 'grpo_loop_time': '419.44253540', 'ref_model_time': '111.95742035', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89250350', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.27685547', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.410e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.01001408', 'kl_moving': '0.01001408', 'kl_stepwise': '-1.897e-08', 'gathered_avg_reward_mean': '9.27624512', 'gathered_avg_reward_std': '0.89328235', 'gathered_video_score2_mq_reward_mean': '3.17871094', 'gathered_video_score2_mq_reward_std': '0.28511444', 'gathered_video_score2_overall_reward_mean': '9.27685547', 'gathered_video_score2_overall_reward_std': '0.89327437', 'gathered_video_score2_ta_reward_mean': '2.96411133', 'gathered_video_score2_ta_reward_std': '0.61871028', 'gathered_video_score2_vq_reward_mean': '3.13342285', 'gathered_video_score2_vq_reward_std': '0.37822935', 'gathered_ori_avg_reward_mean': '9.27624512', 'gathered_ori_avg_reward_std': '0.89328235'} +2026-06-28 10:04:16.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 10:04:16.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:22.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_0/index_4742_seed_14698_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 10:05:38.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:38.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:06:50.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_0/index_4742_seed_14699_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:07:11.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 10:07:11.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.226 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 378 train_steps 11120 {'step': '378/10000', 'learning_rate': '0.00001000', 'loss': '-4.424e-09', 'grad_norm': '0.00548851', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.33419800', 'grpo_loop_time': '423.02899170', 'ref_model_time': '112.96833038', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89110011', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.91992188', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.430e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.424e-09', 'kl_loss': '0.01072347', 'kl_moving': '0.01072347', 'kl_stepwise': '-1.907e-08', 'gathered_avg_reward_mean': '8.91809082', 'gathered_avg_reward_std': '0.93345380', 'gathered_video_score2_mq_reward_mean': '3.12939453', 'gathered_video_score2_mq_reward_std': '0.38665432', 'gathered_video_score2_overall_reward_mean': '8.91992188', 'gathered_video_score2_overall_reward_std': '0.93686301', 'gathered_video_score2_ta_reward_mean': '2.75939941', 'gathered_video_score2_ta_reward_std': '0.51347911', 'gathered_video_score2_vq_reward_mean': '3.02929688', 'gathered_video_score2_vq_reward_std': '0.41256270', 'gathered_ori_avg_reward_mean': '8.91809082', 'gathered_ori_avg_reward_std': '0.93345380'} +2026-06-28 10:18:09.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 10:18:09.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:16.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_0/index_7951_seed_8134_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 10:19:28.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:28.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:20:41.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_0/index_7951_seed_8135_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:21:00.940 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 10:21:02.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:53.945 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 379 train_steps 11160 {'step': '379/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00519483', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.07312012', 'grpo_loop_time': '420.55310059', 'ref_model_time': '112.19379425', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86049181', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.27587891', 'token_credit_video_advantage_mean': '-2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.01043616', 'kl_moving': '0.01043616', 'kl_stepwise': '-1.940e-08', 'gathered_avg_reward_mean': '9.27758789', 'gathered_avg_reward_std': '1.13337982', 'gathered_video_score2_mq_reward_mean': '3.17858887', 'gathered_video_score2_mq_reward_std': '0.42174584', 'gathered_video_score2_overall_reward_mean': '9.27587891', 'gathered_video_score2_overall_reward_std': '1.13273919', 'gathered_video_score2_ta_reward_mean': '3.12707520', 'gathered_video_score2_ta_reward_std': '0.62993693', 'gathered_video_score2_vq_reward_mean': '2.97192383', 'gathered_video_score2_vq_reward_std': '0.43929407', 'gathered_ori_avg_reward_mean': '9.27758789', 'gathered_ori_avg_reward_std': '1.13337982'} +2026-06-28 10:31:54.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 10:31:54.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:33:00.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_0/index_2379_seed_1898_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 10:33:14.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:33:14.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:26.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_0/index_2379_seed_1899_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:34:53.176 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 10:34:54.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:45:43.522 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 380 train_steps 11200 {'step': '380/10000', 'learning_rate': '0.00001000', 'loss': '2.328e-09', 'grad_norm': '0.00479685', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.26174927', 'grpo_loop_time': '419.07302856', 'ref_model_time': '111.45446777', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87388980', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.54492188', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.328e-09', 'kl_loss': '0.01103513', 'kl_moving': '0.01103514', 'kl_stepwise': '-1.946e-08', 'gathered_avg_reward_mean': '9.54211426', 'gathered_avg_reward_std': '1.30472374', 'gathered_video_score2_mq_reward_mean': '3.30590820', 'gathered_video_score2_mq_reward_std': '0.39902878', 'gathered_video_score2_overall_reward_mean': '9.54492188', 'gathered_video_score2_overall_reward_std': '1.30305934', 'gathered_video_score2_ta_reward_mean': '3.01672363', 'gathered_video_score2_ta_reward_std': '0.75813568', 'gathered_video_score2_vq_reward_mean': '3.21948242', 'gathered_video_score2_vq_reward_std': '0.41934854', 'gathered_ori_avg_reward_mean': '9.54211426', 'gathered_ori_avg_reward_std': '1.30472374'} +2026-06-28 10:46:19.002 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 10:46:20.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:27.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_0/index_7408_seed_31574_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 10:47:40.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:40.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:48:51.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_0/index_7408_seed_31575_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:49:12.493 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 10:49:13.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:07.987 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 381 train_steps 11240 {'step': '381/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00553336', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.65267944', 'grpo_loop_time': '420.29861450', 'ref_model_time': '113.08105469', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86264497', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14160156', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.402e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00897697', 'kl_moving': '0.00897697', 'kl_stepwise': '-1.898e-08', 'gathered_avg_reward_mean': '9.14367676', 'gathered_avg_reward_std': '1.03269768', 'gathered_video_score2_mq_reward_mean': '3.19458008', 'gathered_video_score2_mq_reward_std': '0.28223744', 'gathered_video_score2_overall_reward_mean': '9.14160156', 'gathered_video_score2_overall_reward_std': '1.03238821', 'gathered_video_score2_ta_reward_mean': '2.94763184', 'gathered_video_score2_ta_reward_std': '0.68474472', 'gathered_video_score2_vq_reward_mean': '3.00146484', 'gathered_video_score2_vq_reward_std': '0.43737546', 'gathered_ori_avg_reward_mean': '9.14367676', 'gathered_ori_avg_reward_std': '1.03269780'} +2026-06-28 11:00:08.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 11:00:08.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:14.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_0/index_5970_seed_1332_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 11:01:29.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:29.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:39.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_0/index_5970_seed_1333_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 11:02:59.508 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:02:59.675 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 11:03:00.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.334 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 382 train_steps 11280 {'step': '382/10000', 'learning_rate': '0.00001000', 'loss': '6.286e-09', 'grad_norm': '0.00620242', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.97171021', 'grpo_loop_time': '419.26068115', 'ref_model_time': '112.62211609', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86594832', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.35644531', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.403e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.054e-09', 'kl_loss': '0.00922997', 'kl_moving': '0.00922997', 'kl_stepwise': '-1.887e-08', 'gathered_avg_reward_mean': '9.35449219', 'gathered_avg_reward_std': '1.08800852', 'gathered_video_score2_mq_reward_mean': '3.24829102', 'gathered_video_score2_mq_reward_std': '0.35947177', 'gathered_video_score2_overall_reward_mean': '9.35644531', 'gathered_video_score2_overall_reward_std': '1.08562434', 'gathered_video_score2_ta_reward_mean': '3.12902832', 'gathered_video_score2_ta_reward_std': '0.58692622', 'gathered_video_score2_vq_reward_mean': '2.97717285', 'gathered_video_score2_vq_reward_std': '0.40701294', 'gathered_ori_avg_reward_mean': '9.35449219', 'gathered_ori_avg_reward_std': '1.08800852'} +2026-06-28 11:13:51.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 11:13:51.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:57.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_0/index_4668_seed_13690_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 11:15:11.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:15:11.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:21.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_0/index_4668_seed_13691_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:16:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 11:16:43.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.353 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 383 train_steps 11320 {'step': '383/10000', 'learning_rate': '0.00001000', 'loss': '9.662e-09', 'grad_norm': '0.00517556', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.58587646', 'grpo_loop_time': '418.41360474', 'ref_model_time': '111.54280853', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87813032', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.44042969', 'token_credit_video_advantage_mean': '2.969e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.406e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.662e-09', 'kl_loss': '0.00878351', 'kl_moving': '0.00878351', 'kl_stepwise': '-1.823e-08', 'gathered_avg_reward_mean': '9.44677734', 'gathered_avg_reward_std': '1.01536846', 'gathered_video_score2_mq_reward_mean': '3.21484375', 'gathered_video_score2_mq_reward_std': '0.39308795', 'gathered_video_score2_overall_reward_mean': '9.44042969', 'gathered_video_score2_overall_reward_std': '1.01797688', 'gathered_video_score2_ta_reward_mean': '3.10791016', 'gathered_video_score2_ta_reward_std': '0.51674622', 'gathered_video_score2_vq_reward_mean': '3.12402344', 'gathered_video_score2_vq_reward_std': '0.41988894', 'gathered_ori_avg_reward_mean': '9.44677734', 'gathered_ori_avg_reward_std': '1.01536846'} +2026-06-28 11:27:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 11:27:32.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:38.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_0/index_9975_seed_1463_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 11:28:52.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:52.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:03.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_0/index_9975_seed_1464_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:30:22.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 11:30:23.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.539 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 384 train_steps 11360 {'step': '384/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00642227', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.13436890', 'grpo_loop_time': '420.19894409', 'ref_model_time': '111.31213379', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89003044', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.74267578', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.465e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.00965461', 'kl_moving': '0.00965461', 'kl_stepwise': '-1.896e-08', 'gathered_avg_reward_mean': '9.74206543', 'gathered_avg_reward_std': '0.96784937', 'gathered_video_score2_mq_reward_mean': '3.35888672', 'gathered_video_score2_mq_reward_std': '0.38685653', 'gathered_video_score2_overall_reward_mean': '9.74267578', 'gathered_video_score2_overall_reward_std': '0.96913964', 'gathered_video_score2_ta_reward_mean': '3.12976074', 'gathered_video_score2_ta_reward_std': '0.54981184', 'gathered_video_score2_vq_reward_mean': '3.25341797', 'gathered_video_score2_vq_reward_std': '0.36529672', 'gathered_ori_avg_reward_mean': '9.74206543', 'gathered_ori_avg_reward_std': '0.96784931'} +2026-06-28 11:41:15.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 11:41:15.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:21.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_0/index_9340_seed_6470_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 11:42:35.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:35.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:43:47.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_0/index_9340_seed_6471_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:44:07.731 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 11:44:08.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.364 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 385 train_steps 11400 {'step': '385/10000', 'learning_rate': '0.00001000', 'loss': '-6.170e-09', 'grad_norm': '0.00594105', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.32296753', 'grpo_loop_time': '422.23394775', 'ref_model_time': '112.13262939', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87584710', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24414062', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.390e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.033e-09', 'kl_loss': '0.00918827', 'kl_moving': '0.00918827', 'kl_stepwise': '-1.889e-08', 'gathered_avg_reward_mean': '9.24450684', 'gathered_avg_reward_std': '0.90222079', 'gathered_video_score2_mq_reward_mean': '3.28076172', 'gathered_video_score2_mq_reward_std': '0.34483546', 'gathered_video_score2_overall_reward_mean': '9.24414062', 'gathered_video_score2_overall_reward_std': '0.89838445', 'gathered_video_score2_ta_reward_mean': '2.80932617', 'gathered_video_score2_ta_reward_std': '0.56469589', 'gathered_video_score2_vq_reward_mean': '3.15441895', 'gathered_video_score2_vq_reward_std': '0.43699560', 'gathered_ori_avg_reward_mean': '9.24450684', 'gathered_ori_avg_reward_std': '0.90222079'} +2026-06-28 11:55:03.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 11:55:03.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:09.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_0/index_5864_seed_32733_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 11:56:24.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:34.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_0/index_5864_seed_32734_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:57:56.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 11:57:57.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.488 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 386 train_steps 11440 {'step': '386/10000', 'learning_rate': '0.00001000', 'loss': '2.206e-08', 'grad_norm': '0.00791046', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.64187622', 'grpo_loop_time': '419.58770752', 'ref_model_time': '112.80506897', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85938650', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.13769531', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.405e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.154e-08', 'kl_loss': '0.01081998', 'kl_moving': '0.01081998', 'kl_stepwise': '-1.927e-08', 'gathered_avg_reward_mean': '9.13867188', 'gathered_avg_reward_std': '0.91533554', 'gathered_video_score2_mq_reward_mean': '3.11743164', 'gathered_video_score2_mq_reward_std': '0.36996117', 'gathered_video_score2_overall_reward_mean': '9.13769531', 'gathered_video_score2_overall_reward_std': '0.91422474', 'gathered_video_score2_ta_reward_mean': '3.08984375', 'gathered_video_score2_ta_reward_std': '0.56639445', 'gathered_video_score2_vq_reward_mean': '2.93139648', 'gathered_video_score2_vq_reward_std': '0.45749524', 'gathered_ori_avg_reward_mean': '9.13867188', 'gathered_ori_avg_reward_std': '0.91533554'} +2026-06-28 12:08:48.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 12:08:48.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:54.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_0/index_526_seed_28932_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 12:10:12.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:10:12.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:26.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_0/index_526_seed_28933_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:11:47.132 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 12:11:48.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.481 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 387 train_steps 11480 {'step': '387/10000', 'learning_rate': '0.00001000', 'loss': '4.377e-09', 'grad_norm': '0.00960871', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '291.61889648', 'grpo_loop_time': '420.57693481', 'ref_model_time': '111.29083252', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84009314', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.28417969', 'token_credit_video_advantage_mean': '-1.746e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.446e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.377e-09', 'kl_loss': '0.01068034', 'kl_moving': '0.01068034', 'kl_stepwise': '-1.935e-08', 'gathered_avg_reward_mean': '9.28295898', 'gathered_avg_reward_std': '0.79211491', 'gathered_video_score2_mq_reward_mean': '3.14135742', 'gathered_video_score2_mq_reward_std': '0.24279206', 'gathered_video_score2_overall_reward_mean': '9.28417969', 'gathered_video_score2_overall_reward_std': '0.79046750', 'gathered_video_score2_ta_reward_mean': '3.02148438', 'gathered_video_score2_ta_reward_std': '0.57742351', 'gathered_video_score2_vq_reward_mean': '3.12011719', 'gathered_video_score2_vq_reward_std': '0.40773210', 'gathered_ori_avg_reward_mean': '9.28295898', 'gathered_ori_avg_reward_std': '0.79211497'} +2026-06-28 12:22:40.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 12:22:40.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:46.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_0/index_3313_seed_16288_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 12:23:59.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:59.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:14.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_0/index_3313_seed_16289_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:25:34.048 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 12:25:34.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.334 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 388 train_steps 11520 {'step': '388/10000', 'learning_rate': '0.00001000', 'loss': '-4.657e-09', 'grad_norm': '0.02670418', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.50027466', 'grpo_loop_time': '418.55810547', 'ref_model_time': '112.33090973', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86098069', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.86132812', 'token_credit_video_advantage_mean': '1.024e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.422e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-4.657e-09', 'kl_loss': '0.01167334', 'kl_moving': '0.01167335', 'kl_stepwise': '-1.961e-08', 'gathered_avg_reward_mean': '9.86010742', 'gathered_avg_reward_std': '0.63973969', 'gathered_video_score2_mq_reward_mean': '3.31396484', 'gathered_video_score2_mq_reward_std': '0.30450866', 'gathered_video_score2_overall_reward_mean': '9.86132812', 'gathered_video_score2_overall_reward_std': '0.64133620', 'gathered_video_score2_ta_reward_mean': '3.35351562', 'gathered_video_score2_ta_reward_std': '0.42365709', 'gathered_video_score2_vq_reward_mean': '3.19262695', 'gathered_video_score2_vq_reward_std': '0.33803234', 'gathered_ori_avg_reward_mean': '9.86010742', 'gathered_ori_avg_reward_std': '0.63973969'} +2026-06-28 12:36:24.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 12:36:24.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:30.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_0/index_4172_seed_32132_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 12:37:48.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:48.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:38:55.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_0/index_4172_seed_32133_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:39:16.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 12:39:17.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.661 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 389 train_steps 11560 {'step': '389/10000', 'learning_rate': '0.00001000', 'loss': '3.166e-08', 'grad_norm': '0.00655730', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.01861572', 'grpo_loop_time': '420.09356689', 'ref_model_time': '112.17794037', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85643291', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.56933594', 'token_credit_video_advantage_mean': '-7.451e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.460e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.166e-08', 'kl_loss': '0.01122696', 'kl_moving': '0.01122696', 'kl_stepwise': '-1.989e-08', 'gathered_avg_reward_mean': '9.56958008', 'gathered_avg_reward_std': '1.29043984', 'gathered_video_score2_mq_reward_mean': '3.25244141', 'gathered_video_score2_mq_reward_std': '0.43421987', 'gathered_video_score2_overall_reward_mean': '9.56933594', 'gathered_video_score2_overall_reward_std': '1.29301560', 'gathered_video_score2_ta_reward_mean': '3.23681641', 'gathered_video_score2_ta_reward_std': '0.60093099', 'gathered_video_score2_vq_reward_mean': '3.08032227', 'gathered_video_score2_vq_reward_std': '0.51334774', 'gathered_ori_avg_reward_mean': '9.56958008', 'gathered_ori_avg_reward_std': '1.29043984'} +2026-06-28 12:50:08.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 12:50:08.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_0/index_8860_seed_2783_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 12:51:34.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:34.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:52:41.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_0/index_8860_seed_2784_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:53:04.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 12:53:04.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:03:55.316 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 390 train_steps 11600 {'step': '390/10000', 'learning_rate': '0.00001000', 'loss': '2.328e-10', 'grad_norm': '0.00563865', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.99627686', 'grpo_loop_time': '418.93408203', 'ref_model_time': '112.00424194', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88339448', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.32568359', 'token_credit_video_advantage_mean': '3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.439e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.164e-10', 'kl_loss': '0.01227954', 'kl_moving': '0.01227954', 'kl_stepwise': '-1.986e-08', 'gathered_avg_reward_mean': '9.32324219', 'gathered_avg_reward_std': '1.07697833', 'gathered_video_score2_mq_reward_mean': '3.25927734', 'gathered_video_score2_mq_reward_std': '0.31496486', 'gathered_video_score2_overall_reward_mean': '9.32568359', 'gathered_video_score2_overall_reward_std': '1.07966387', 'gathered_video_score2_ta_reward_mean': '2.95593262', 'gathered_video_score2_ta_reward_std': '0.58812803', 'gathered_video_score2_vq_reward_mean': '3.10803223', 'gathered_video_score2_vq_reward_std': '0.45055330', 'gathered_ori_avg_reward_mean': '9.32324219', 'gathered_ori_avg_reward_std': '1.07697845'} +2026-06-28 13:04:30.931 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 13:04:33.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:40.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_0/index_7719_seed_23775_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 13:05:56.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:56.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:08.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_0/index_7719_seed_23776_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 13:07:33.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:07:33.182 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 13:07:33.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.433 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 391 train_steps 11640 {'step': '391/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00582683', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '293.30654907', 'grpo_loop_time': '418.13351440', 'ref_model_time': '111.61324310', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86610043', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.20605469', 'token_credit_video_advantage_mean': '1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.463e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.00922614', 'kl_moving': '0.00922615', 'kl_stepwise': '-1.879e-08', 'gathered_avg_reward_mean': '9.20996094', 'gathered_avg_reward_std': '1.20790040', 'gathered_video_score2_mq_reward_mean': '3.18457031', 'gathered_video_score2_mq_reward_std': '0.36449599', 'gathered_video_score2_overall_reward_mean': '9.20605469', 'gathered_video_score2_overall_reward_std': '1.20914328', 'gathered_video_score2_ta_reward_mean': '2.87719727', 'gathered_video_score2_ta_reward_std': '0.72465736', 'gathered_video_score2_vq_reward_mean': '3.14819336', 'gathered_video_score2_vq_reward_std': '0.45572725', 'gathered_ori_avg_reward_mean': '9.20996094', 'gathered_ori_avg_reward_std': '1.20790040'} +2026-06-28 13:18:22.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 13:18:22.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:29.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_0/index_3072_seed_2586_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 13:19:48.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:48.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:20:54.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_0/index_3072_seed_2587_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:21:13.466 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 13:21:14.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:06.952 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 392 train_steps 11680 {'step': '392/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.00549130', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.78048706', 'grpo_loop_time': '419.22775269', 'ref_model_time': '113.24245453', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85079300', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.05029297', 'token_credit_video_advantage_mean': '6.054e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.476e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00919531', 'kl_moving': '0.00919531', 'kl_stepwise': '-2.026e-08', 'gathered_avg_reward_mean': '9.04858398', 'gathered_avg_reward_std': '1.06093216', 'gathered_video_score2_mq_reward_mean': '3.11413574', 'gathered_video_score2_mq_reward_std': '0.37252659', 'gathered_video_score2_overall_reward_mean': '9.05029297', 'gathered_video_score2_overall_reward_std': '1.05989194', 'gathered_video_score2_ta_reward_mean': '3.00463867', 'gathered_video_score2_ta_reward_std': '0.46218285', 'gathered_video_score2_vq_reward_mean': '2.92980957', 'gathered_video_score2_vq_reward_std': '0.55605179', 'gathered_ori_avg_reward_mean': '9.04858398', 'gathered_ori_avg_reward_std': '1.06093216'} +2026-06-28 13:32:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 13:32:07.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:13.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_0/index_9710_seed_29807_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 13:33:30.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:30.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:38.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_0/index_9710_seed_29808_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 13:34:59.611 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:34:59.776 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 13:35:00.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.743 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 393 train_steps 11720 {'step': '393/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00588697', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.86364746', 'grpo_loop_time': '417.86010742', 'ref_model_time': '112.41044617', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87490499', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.18603516', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.454e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.01004692', 'kl_moving': '0.01004692', 'kl_stepwise': '-1.999e-08', 'gathered_avg_reward_mean': '9.18725586', 'gathered_avg_reward_std': '0.88737458', 'gathered_video_score2_mq_reward_mean': '3.15234375', 'gathered_video_score2_mq_reward_std': '0.30119950', 'gathered_video_score2_overall_reward_mean': '9.18603516', 'gathered_video_score2_overall_reward_std': '0.88805878', 'gathered_video_score2_ta_reward_mean': '3.08007812', 'gathered_video_score2_ta_reward_std': '0.44951162', 'gathered_video_score2_vq_reward_mean': '2.95483398', 'gathered_video_score2_vq_reward_std': '0.39491895', 'gathered_ori_avg_reward_mean': '9.18725586', 'gathered_ori_avg_reward_std': '0.88737452'} +2026-06-28 13:45:49.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 13:45:49.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:49.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_0/index_1173_seed_28053_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 13:47:11.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:47:11.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:22.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_0/index_1173_seed_28054_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 13:48:41.544 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:48:41.715 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 13:48:42.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.739 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 394 train_steps 11760 {'step': '394/10000', 'learning_rate': '0.00001000', 'loss': '2.014e-08', 'grad_norm': '0.00521249', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.37609863', 'grpo_loop_time': '417.76171875', 'ref_model_time': '111.73397827', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87014294', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.50732422', 'token_credit_video_advantage_mean': '7.334e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.438e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.014e-08', 'kl_loss': '0.00928765', 'kl_moving': '0.00928766', 'kl_stepwise': '-1.910e-08', 'gathered_avg_reward_mean': '9.50317383', 'gathered_avg_reward_std': '0.92179602', 'gathered_video_score2_mq_reward_mean': '3.25048828', 'gathered_video_score2_mq_reward_std': '0.27090019', 'gathered_video_score2_overall_reward_mean': '9.50732422', 'gathered_video_score2_overall_reward_std': '0.91920203', 'gathered_video_score2_ta_reward_mean': '3.05322266', 'gathered_video_score2_ta_reward_std': '0.66844529', 'gathered_video_score2_vq_reward_mean': '3.19946289', 'gathered_video_score2_vq_reward_std': '0.36350948', 'gathered_ori_avg_reward_mean': '9.50317383', 'gathered_ori_avg_reward_std': '0.92179608'} +2026-06-28 13:59:30.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 13:59:30.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:30.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_0/index_9080_seed_18289_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 14:00:53.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:53.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:00.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_0/index_9080_seed_18290_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:02:20.412 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 14:02:21.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.241 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 395 train_steps 11800 {'step': '395/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-08', 'grad_norm': '0.00502280', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.42547607', 'grpo_loop_time': '417.31109619', 'ref_model_time': '111.17123413', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88779372', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.35839844', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.426e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.608e-08', 'kl_loss': '0.00964090', 'kl_moving': '0.00964091', 'kl_stepwise': '-1.880e-08', 'gathered_avg_reward_mean': '9.35900879', 'gathered_avg_reward_std': '1.29738486', 'gathered_video_score2_mq_reward_mean': '3.20861816', 'gathered_video_score2_mq_reward_std': '0.38435069', 'gathered_video_score2_overall_reward_mean': '9.35839844', 'gathered_video_score2_overall_reward_std': '1.29976583', 'gathered_video_score2_ta_reward_mean': '2.97229004', 'gathered_video_score2_ta_reward_std': '0.75193655', 'gathered_video_score2_vq_reward_mean': '3.17810059', 'gathered_video_score2_vq_reward_std': '0.45520005', 'gathered_ori_avg_reward_mean': '9.35900879', 'gathered_ori_avg_reward_std': '1.29738486'} +2026-06-28 14:13:08.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 14:13:08.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_0/index_8064_seed_13841_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 14:14:33.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:33.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:15:42.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_0/index_8064_seed_13842_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:16:08.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 14:16:09.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.047 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 396 train_steps 11840 {'step': '396/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-09', 'grad_norm': '0.01034704', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '292.97869873', 'grpo_loop_time': '420.75531006', 'ref_model_time': '110.74285126', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86304688', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.99804688', 'token_credit_video_advantage_mean': '9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.453e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.01113715', 'kl_moving': '0.01113716', 'kl_stepwise': '-2.050e-08', 'gathered_avg_reward_mean': '8.99633789', 'gathered_avg_reward_std': '1.19649744', 'gathered_video_score2_mq_reward_mean': '3.18933105', 'gathered_video_score2_mq_reward_std': '0.41645724', 'gathered_video_score2_overall_reward_mean': '8.99804688', 'gathered_video_score2_overall_reward_std': '1.20019019', 'gathered_video_score2_ta_reward_mean': '2.83483887', 'gathered_video_score2_ta_reward_std': '0.66802853', 'gathered_video_score2_vq_reward_mean': '2.97216797', 'gathered_video_score2_vq_reward_std': '0.51167750', 'gathered_ori_avg_reward_mean': '8.99633789', 'gathered_ori_avg_reward_std': '1.19649756'} +2026-06-28 14:27:02.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 14:27:02.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:08.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_0/index_8894_seed_24509_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 14:28:23.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:23.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:32.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_0/index_8894_seed_24510_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:29:50.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 14:29:51.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.829 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 397 train_steps 11880 {'step': '397/10000', 'learning_rate': '0.00001000', 'loss': '1.630e-08', 'grad_norm': '0.00452608', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '281.56494141', 'grpo_loop_time': '420.64767456', 'ref_model_time': '111.19776917', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89019924', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.68261719', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.436e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.779e-09', 'kl_loss': '0.01117512', 'kl_moving': '0.01117512', 'kl_stepwise': '-2.015e-08', 'gathered_avg_reward_mean': '9.68005371', 'gathered_avg_reward_std': '1.29421353', 'gathered_video_score2_mq_reward_mean': '3.34155273', 'gathered_video_score2_mq_reward_std': '0.47719339', 'gathered_video_score2_overall_reward_mean': '9.68261719', 'gathered_video_score2_overall_reward_std': '1.29199731', 'gathered_video_score2_ta_reward_mean': '3.11743164', 'gathered_video_score2_ta_reward_std': '0.65004003', 'gathered_video_score2_vq_reward_mean': '3.22106934', 'gathered_video_score2_vq_reward_std': '0.47125730', 'gathered_ori_avg_reward_mean': '9.68005371', 'gathered_ori_avg_reward_std': '1.29421353'} +2026-06-28 14:40:43.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 14:40:43.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:50.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_0/index_1626_seed_24058_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 14:42:04.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:42:04.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:15.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_0/index_1626_seed_24059_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:43:35.144 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 14:43:35.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.180 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 398 train_steps 11920 {'step': '398/10000', 'learning_rate': '0.00001000', 'loss': '4.983e-09', 'grad_norm': '0.00529853', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.18872070', 'grpo_loop_time': '419.03320312', 'ref_model_time': '111.25766754', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88698882', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.25146484', 'token_credit_video_advantage_mean': '-6.985e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.427e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.054e-10', 'kl_loss': '0.01226212', 'kl_moving': '0.01226212', 'kl_stepwise': '-1.893e-08', 'gathered_avg_reward_mean': '9.25366211', 'gathered_avg_reward_std': '1.10232925', 'gathered_video_score2_mq_reward_mean': '3.23730469', 'gathered_video_score2_mq_reward_std': '0.36194390', 'gathered_video_score2_overall_reward_mean': '9.25146484', 'gathered_video_score2_overall_reward_std': '1.10588181', 'gathered_video_score2_ta_reward_mean': '2.95996094', 'gathered_video_score2_ta_reward_std': '0.57852817', 'gathered_video_score2_vq_reward_mean': '3.05639648', 'gathered_video_score2_vq_reward_std': '0.45733887', 'gathered_ori_avg_reward_mean': '9.25366211', 'gathered_ori_avg_reward_std': '1.10232925'} +2026-06-28 14:54:25.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 14:54:25.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:31.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_0/index_9447_seed_29206_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 14:55:47.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:47.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:56:55.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_0/index_9447_seed_29207_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:57:14.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 14:57:15.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.127 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 399 train_steps 11960 {'step': '399/10000', 'learning_rate': '0.00001000', 'loss': '-1.118e-08', 'grad_norm': '0.00524384', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.98800659', 'grpo_loop_time': '419.35543823', 'ref_model_time': '112.60072327', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86271709', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.82861328', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.448e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.118e-08', 'kl_loss': '0.01298583', 'kl_moving': '0.01298584', 'kl_stepwise': '-1.967e-08', 'gathered_avg_reward_mean': '9.82971191', 'gathered_avg_reward_std': '0.87313992', 'gathered_video_score2_mq_reward_mean': '3.29614258', 'gathered_video_score2_mq_reward_std': '0.39434198', 'gathered_video_score2_overall_reward_mean': '9.82861328', 'gathered_video_score2_overall_reward_std': '0.87300962', 'gathered_video_score2_ta_reward_mean': '3.32360840', 'gathered_video_score2_ta_reward_std': '0.51166993', 'gathered_video_score2_vq_reward_mean': '3.20996094', 'gathered_video_score2_vq_reward_std': '0.39590067', 'gathered_ori_avg_reward_mean': '9.82971191', 'gathered_ori_avg_reward_std': '0.87313992'} +2026-06-28 15:08:07.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 15:08:07.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:13.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_0/index_4227_seed_27751_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 15:09:31.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:31.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:10:39.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_0/index_4227_seed_27752_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:11:01.175 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 15:11:02.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:21:50.505 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 400 train_steps 12000 {'step': '400/10000', 'learning_rate': '0.00001000', 'loss': '-9.313e-09', 'grad_norm': '0.00519590', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.35604858', 'grpo_loop_time': '418.06430054', 'ref_model_time': '111.55927277', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87183380', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.62255859', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.386e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.676e-08', 'kl_loss': '0.01259954', 'kl_moving': '0.01259954', 'kl_stepwise': '-1.896e-08', 'gathered_avg_reward_mean': '9.61999512', 'gathered_avg_reward_std': '1.10900652', 'gathered_video_score2_mq_reward_mean': '3.28552246', 'gathered_video_score2_mq_reward_std': '0.44787276', 'gathered_video_score2_overall_reward_mean': '9.62255859', 'gathered_video_score2_overall_reward_std': '1.10959923', 'gathered_video_score2_ta_reward_mean': '3.08984375', 'gathered_video_score2_ta_reward_std': '0.51385933', 'gathered_video_score2_vq_reward_mean': '3.24462891', 'gathered_video_score2_vq_reward_std': '0.40435082', 'gathered_ori_avg_reward_mean': '9.61999512', 'gathered_ori_avg_reward_std': '1.10900652'} +2026-06-28 15:22:55.961 | INFO | __main__:save_checkpoint:1967 - Checkpoint saved at step 400 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-400 +2026-06-28 15:22:56.093 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 400: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/checkpoint-400 +2026-06-28 15:22:58.855 | INFO | __main__:_prune_old_timer_checkpoints:1799 - Removed stale timer checkpoint: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-375 +2026-06-28 15:23:34.005 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 15:23:36.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:42.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_0/index_1247_seed_6294_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 15:24:59.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:59.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:08.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_0/index_1247_seed_6295_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 15:26:31.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:26:31.197 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 15:26:32.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.059 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 401 train_steps 12040 {'step': '401/10000', 'learning_rate': '0.00001000', 'loss': '-6.519e-09', 'grad_norm': '0.00485159', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.11801147', 'grpo_loop_time': '418.99255371', 'ref_model_time': '111.31457520', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87719804', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.70361328', 'token_credit_video_advantage_mean': '2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.458e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.519e-09', 'kl_loss': '0.00892909', 'kl_moving': '0.00892909', 'kl_stepwise': '-1.958e-08', 'gathered_avg_reward_mean': '9.70324707', 'gathered_avg_reward_std': '1.01818943', 'gathered_video_score2_mq_reward_mean': '3.35424805', 'gathered_video_score2_mq_reward_std': '0.36448741', 'gathered_video_score2_overall_reward_mean': '9.70361328', 'gathered_video_score2_overall_reward_std': '1.01360965', 'gathered_video_score2_ta_reward_mean': '3.17126465', 'gathered_video_score2_ta_reward_std': '0.47378945', 'gathered_video_score2_vq_reward_mean': '3.17773438', 'gathered_video_score2_vq_reward_std': '0.40514517', 'gathered_ori_avg_reward_mean': '9.70324707', 'gathered_ori_avg_reward_std': '1.01818943'} +2026-06-28 15:37:21.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 15:37:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:27.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_0/index_468_seed_18240_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 15:38:41.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:41.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:39:52.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_0/index_468_seed_18241_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:40:14.126 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 15:40:15.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.160 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 402 train_steps 12080 {'step': '402/10000', 'learning_rate': '0.00001000', 'loss': '4.773e-09', 'grad_norm': '0.00613536', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.25363159', 'grpo_loop_time': '420.96478271', 'ref_model_time': '111.63725281', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87992561', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.36181641', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.449e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.773e-09', 'kl_loss': '0.00954045', 'kl_moving': '0.00954045', 'kl_stepwise': '-1.921e-08', 'gathered_avg_reward_mean': '9.36413574', 'gathered_avg_reward_std': '1.20967329', 'gathered_video_score2_mq_reward_mean': '3.20349121', 'gathered_video_score2_mq_reward_std': '0.49067077', 'gathered_video_score2_overall_reward_mean': '9.36181641', 'gathered_video_score2_overall_reward_std': '1.21133816', 'gathered_video_score2_ta_reward_mean': '2.93530273', 'gathered_video_score2_ta_reward_std': '0.66607928', 'gathered_video_score2_vq_reward_mean': '3.22534180', 'gathered_video_score2_vq_reward_std': '0.44771150', 'gathered_ori_avg_reward_mean': '9.36413574', 'gathered_ori_avg_reward_std': '1.20967329'} +2026-06-28 15:51:08.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 15:51:08.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:14.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_0/index_7363_seed_3362_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 15:52:29.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:29.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:53:41.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_0/index_7363_seed_3363_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:54:03.712 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 15:54:04.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.280 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 403 train_steps 12120 {'step': '403/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00584948', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.19836426', 'grpo_loop_time': '418.44091797', 'ref_model_time': '111.63720703', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87636870', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.11718750', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.358e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00994004', 'kl_moving': '0.00994004', 'kl_stepwise': '-1.984e-08', 'gathered_avg_reward_mean': '9.12145996', 'gathered_avg_reward_std': '1.05314648', 'gathered_video_score2_mq_reward_mean': '3.16601562', 'gathered_video_score2_mq_reward_std': '0.42289102', 'gathered_video_score2_overall_reward_mean': '9.11718750', 'gathered_video_score2_overall_reward_std': '1.05337012', 'gathered_video_score2_ta_reward_mean': '2.86975098', 'gathered_video_score2_ta_reward_std': '0.56123900', 'gathered_video_score2_vq_reward_mean': '3.08569336', 'gathered_video_score2_vq_reward_std': '0.39414424', 'gathered_ori_avg_reward_mean': '9.12145996', 'gathered_ori_avg_reward_std': '1.05314648'} +2026-06-28 16:04:54.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 16:04:54.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:06:00.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_0/index_5101_seed_16601_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 16:06:17.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:06:17.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:24.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_0/index_5101_seed_16602_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 16:07:49.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:07:49.431 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 16:07:50.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.670 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 404 train_steps 12160 {'step': '404/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00546034', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.21218872', 'grpo_loop_time': '418.89913940', 'ref_model_time': '111.47734070', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87572968', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.19335938', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.438e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.01085293', 'kl_moving': '0.01085293', 'kl_stepwise': '-1.928e-08', 'gathered_avg_reward_mean': '9.19360352', 'gathered_avg_reward_std': '0.90279853', 'gathered_video_score2_mq_reward_mean': '3.18237305', 'gathered_video_score2_mq_reward_std': '0.28388327', 'gathered_video_score2_overall_reward_mean': '9.19335938', 'gathered_video_score2_overall_reward_std': '0.90199488', 'gathered_video_score2_ta_reward_mean': '2.94702148', 'gathered_video_score2_ta_reward_std': '0.53642797', 'gathered_video_score2_vq_reward_mean': '3.06420898', 'gathered_video_score2_vq_reward_std': '0.39586446', 'gathered_ori_avg_reward_mean': '9.19360352', 'gathered_ori_avg_reward_std': '0.90279859'} +2026-06-28 16:18:40.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 16:18:40.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_0/index_4382_seed_8608_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 16:20:05.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:20:05.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:12.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_0/index_4382_seed_8609_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:21:33.543 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 16:21:34.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.899 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 405 train_steps 12200 {'step': '405/10000', 'learning_rate': '0.00001000', 'loss': '-6.054e-09', 'grad_norm': '0.00618931', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.71813965', 'grpo_loop_time': '418.60687256', 'ref_model_time': '111.22199249', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86468709', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.39941406', 'token_credit_video_advantage_mean': '6.752e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.459e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-6.054e-09', 'kl_loss': '0.01089983', 'kl_moving': '0.01089983', 'kl_stepwise': '-1.909e-08', 'gathered_avg_reward_mean': '9.39526367', 'gathered_avg_reward_std': '0.91941756', 'gathered_video_score2_mq_reward_mean': '3.27978516', 'gathered_video_score2_mq_reward_std': '0.33168316', 'gathered_video_score2_overall_reward_mean': '9.39941406', 'gathered_video_score2_overall_reward_std': '0.91813445', 'gathered_video_score2_ta_reward_mean': '2.91674805', 'gathered_video_score2_ta_reward_std': '0.60395670', 'gathered_video_score2_vq_reward_mean': '3.19873047', 'gathered_video_score2_vq_reward_std': '0.44511944', 'gathered_ori_avg_reward_mean': '9.39526367', 'gathered_ori_avg_reward_std': '0.91941762'} +2026-06-28 16:32:22.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 16:32:23.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_0/index_9280_seed_7262_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 16:33:44.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:44.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:34:53.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_0/index_9280_seed_7263_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:35:13.594 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 16:35:14.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.105 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 406 train_steps 12240 {'step': '406/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.01247960', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.20898438', 'grpo_loop_time': '420.29885864', 'ref_model_time': '111.88780975', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87016851', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.10351562', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.425e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.01218728', 'kl_moving': '0.01218728', 'kl_stepwise': '-1.967e-08', 'gathered_avg_reward_mean': '9.10119629', 'gathered_avg_reward_std': '1.18597269', 'gathered_video_score2_mq_reward_mean': '3.13464355', 'gathered_video_score2_mq_reward_std': '0.41151130', 'gathered_video_score2_overall_reward_mean': '9.10351562', 'gathered_video_score2_overall_reward_std': '1.18598115', 'gathered_video_score2_ta_reward_mean': '2.99816895', 'gathered_video_score2_ta_reward_std': '0.61366951', 'gathered_video_score2_vq_reward_mean': '2.96838379', 'gathered_video_score2_vq_reward_std': '0.37249026', 'gathered_ori_avg_reward_mean': '9.10119629', 'gathered_ori_avg_reward_std': '1.18597269'} +2026-06-28 16:46:07.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 16:46:07.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:13.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_0/index_9440_seed_2500_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 16:47:30.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:30.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:39.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_0/index_9440_seed_2501_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:48:59.471 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 16:49:00.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.103 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 407 train_steps 12280 {'step': '407/10000', 'learning_rate': '0.00001000', 'loss': '-1.257e-08', 'grad_norm': '0.00595535', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.43084717', 'grpo_loop_time': '418.50933838', 'ref_model_time': '112.47563934', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89906752', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.59326172', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.400e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.257e-08', 'kl_loss': '0.01151382', 'kl_moving': '0.01151383', 'kl_stepwise': '-1.833e-08', 'gathered_avg_reward_mean': '9.59130859', 'gathered_avg_reward_std': '1.02546680', 'gathered_video_score2_mq_reward_mean': '3.30981445', 'gathered_video_score2_mq_reward_std': '0.36562118', 'gathered_video_score2_overall_reward_mean': '9.59326172', 'gathered_video_score2_overall_reward_std': '1.02044368', 'gathered_video_score2_ta_reward_mean': '3.00610352', 'gathered_video_score2_ta_reward_std': '0.54692292', 'gathered_video_score2_vq_reward_mean': '3.27539062', 'gathered_video_score2_vq_reward_std': '0.31162599', 'gathered_ori_avg_reward_mean': '9.59130859', 'gathered_ori_avg_reward_std': '1.02546680'} +2026-06-28 16:59:50.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 16:59:50.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:56.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_0/index_4734_seed_18126_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 17:01:14.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:01:14.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:21.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_0/index_4734_seed_18127_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 17:02:43.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:02:43.833 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 17:02:44.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.362 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 408 train_steps 12320 {'step': '408/10000', 'learning_rate': '0.00001000', 'loss': '-1.211e-08', 'grad_norm': '0.00603748', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.85244751', 'grpo_loop_time': '418.75802612', 'ref_model_time': '111.49743652', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88741320', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '10.01953125', 'token_credit_video_advantage_mean': '-9.313e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.403e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.211e-08', 'kl_loss': '0.01227837', 'kl_moving': '0.01227837', 'kl_stepwise': '-1.927e-08', 'gathered_avg_reward_mean': '10.01611328', 'gathered_avg_reward_std': '1.18492472', 'gathered_video_score2_mq_reward_mean': '3.46191406', 'gathered_video_score2_mq_reward_std': '0.42630839', 'gathered_video_score2_overall_reward_mean': '10.01953125', 'gathered_video_score2_overall_reward_std': '1.18512690', 'gathered_video_score2_ta_reward_mean': '3.25146484', 'gathered_video_score2_ta_reward_std': '0.60698193', 'gathered_video_score2_vq_reward_mean': '3.30273438', 'gathered_video_score2_vq_reward_std': '0.40805358', 'gathered_ori_avg_reward_mean': '10.01611328', 'gathered_ori_avg_reward_std': '1.18492484'} +2026-06-28 17:13:34.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 17:13:34.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:40.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_0/index_5001_seed_17648_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 17:14:54.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:54.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:08.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_0/index_5001_seed_17649_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:16:31.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 17:16:32.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:20.956 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 409 train_steps 12360 {'step': '409/10000', 'learning_rate': '0.00001000', 'loss': '3.725e-09', 'grad_norm': '0.00577672', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.29141235', 'grpo_loop_time': '419.45788574', 'ref_model_time': '110.92353058', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89577097', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.39843750', 'token_credit_video_advantage_mean': '1.304e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.425e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-3.725e-09', 'kl_loss': '0.01367191', 'kl_moving': '0.01367191', 'kl_stepwise': '-1.992e-08', 'gathered_avg_reward_mean': '9.40002441', 'gathered_avg_reward_std': '0.92645282', 'gathered_video_score2_mq_reward_mean': '3.21508789', 'gathered_video_score2_mq_reward_std': '0.30944234', 'gathered_video_score2_overall_reward_mean': '9.39843750', 'gathered_video_score2_overall_reward_std': '0.92791325', 'gathered_video_score2_ta_reward_mean': '3.00939941', 'gathered_video_score2_ta_reward_std': '0.62273788', 'gathered_video_score2_vq_reward_mean': '3.17553711', 'gathered_video_score2_vq_reward_std': '0.34045768', 'gathered_ori_avg_reward_mean': '9.40002441', 'gathered_ori_avg_reward_std': '0.92645282'} +2026-06-28 17:27:21.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 17:27:21.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_0/index_1821_seed_6329_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 17:28:46.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:46.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:29:53.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_0/index_1821_seed_6330_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 17:30:17.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:30:17.795 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 17:30:18.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:07.390 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 410 train_steps 12400 {'step': '410/10000', 'learning_rate': '0.00001000', 'loss': '-6.985e-09', 'grad_norm': '0.00760895', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.31890869', 'grpo_loop_time': '418.18750000', 'ref_model_time': '111.54270172', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88205087', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.10742188', 'token_credit_video_advantage_mean': '1.211e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.500e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-7.451e-09', 'kl_loss': '0.01647686', 'kl_moving': '0.01647686', 'kl_stepwise': '-2.040e-08', 'gathered_avg_reward_mean': '9.11364746', 'gathered_avg_reward_std': '1.38085938', 'gathered_video_score2_mq_reward_mean': '3.23474121', 'gathered_video_score2_mq_reward_std': '0.49767607', 'gathered_video_score2_overall_reward_mean': '9.10742188', 'gathered_video_score2_overall_reward_std': '1.37895465', 'gathered_video_score2_ta_reward_mean': '2.76367188', 'gathered_video_score2_ta_reward_std': '0.72299534', 'gathered_video_score2_vq_reward_mean': '3.11523438', 'gathered_video_score2_vq_reward_std': '0.47960654', 'gathered_ori_avg_reward_mean': '9.11364746', 'gathered_ori_avg_reward_std': '1.38085938'} +2026-06-28 17:41:42.601 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 17:41:44.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_0/index_7571_seed_18038_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 17:43:09.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:43:09.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:21.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_0/index_7571_seed_18039_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:44:42.579 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 17:44:43.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.576 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 411 train_steps 12440 {'step': '411/10000', 'learning_rate': '0.00001000', 'loss': '-7.451e-09', 'grad_norm': '0.00633748', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.72570801', 'grpo_loop_time': '418.36425781', 'ref_model_time': '111.04081726', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87555164', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.25537109', 'token_credit_video_advantage_mean': '-4.657e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.461e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-9.313e-09', 'kl_loss': '0.00899942', 'kl_moving': '0.00899943', 'kl_stepwise': '-2.032e-08', 'gathered_avg_reward_mean': '9.25903320', 'gathered_avg_reward_std': '0.70283228', 'gathered_video_score2_mq_reward_mean': '3.12939453', 'gathered_video_score2_mq_reward_std': '0.26122954', 'gathered_video_score2_overall_reward_mean': '9.25537109', 'gathered_video_score2_overall_reward_std': '0.70359260', 'gathered_video_score2_ta_reward_mean': '3.03771973', 'gathered_video_score2_ta_reward_std': '0.46085173', 'gathered_video_score2_vq_reward_mean': '3.09191895', 'gathered_video_score2_vq_reward_std': '0.38339686', 'gathered_ori_avg_reward_mean': '9.25903320', 'gathered_ori_avg_reward_std': '0.70283228'} +2026-06-28 17:55:31.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 17:55:31.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:37.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_0/index_1763_seed_19889_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 17:56:54.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:54.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:03.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_0/index_1763_seed_19890_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:58:24.766 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 17:58:25.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.731 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 412 train_steps 12480 {'step': '412/10000', 'learning_rate': '0.00001000', 'loss': '-1.402e-08', 'grad_norm': '0.00612994', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.02447510', 'grpo_loop_time': '418.82522583', 'ref_model_time': '111.31122589', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85741454', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.61523438', 'token_credit_video_advantage_mean': '9.779e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.414e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.774e-08', 'kl_loss': '0.00960745', 'kl_moving': '0.00960746', 'kl_stepwise': '-1.914e-08', 'gathered_avg_reward_mean': '9.61230469', 'gathered_avg_reward_std': '1.06392395', 'gathered_video_score2_mq_reward_mean': '3.25854492', 'gathered_video_score2_mq_reward_std': '0.37958953', 'gathered_video_score2_overall_reward_mean': '9.61523438', 'gathered_video_score2_overall_reward_std': '1.06222534', 'gathered_video_score2_ta_reward_mean': '3.20312500', 'gathered_video_score2_ta_reward_std': '0.53576183', 'gathered_video_score2_vq_reward_mean': '3.15063477', 'gathered_video_score2_vq_reward_std': '0.42434865', 'gathered_ori_avg_reward_mean': '9.61230469', 'gathered_ori_avg_reward_std': '1.06392395'} +2026-06-28 18:09:15.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 18:09:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:15.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:22.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_0/index_1533_seed_5991_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 18:10:40.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:40.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:11:49.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_0/index_1533_seed_5992_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:12:09.445 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 18:12:10.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.010 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 413 train_steps 12520 {'step': '413/10000', 'learning_rate': '0.00001000', 'loss': '8.848e-09', 'grad_norm': '0.00547201', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.38293457', 'grpo_loop_time': '421.26599121', 'ref_model_time': '111.80432892', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87837261', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.51953125', 'token_credit_video_advantage_mean': '-5.821e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.383e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.122e-09', 'kl_loss': '0.00997764', 'kl_moving': '0.00997764', 'kl_stepwise': '-1.828e-08', 'gathered_avg_reward_mean': '9.52160645', 'gathered_avg_reward_std': '1.23067009', 'gathered_video_score2_mq_reward_mean': '3.29589844', 'gathered_video_score2_mq_reward_std': '0.36144820', 'gathered_video_score2_overall_reward_mean': '9.51953125', 'gathered_video_score2_overall_reward_std': '1.23165870', 'gathered_video_score2_ta_reward_mean': '3.20556641', 'gathered_video_score2_ta_reward_std': '0.52971268', 'gathered_video_score2_vq_reward_mean': '3.02014160', 'gathered_video_score2_vq_reward_std': '0.56284881', 'gathered_ori_avg_reward_mean': '9.52160645', 'gathered_ori_avg_reward_std': '1.23067009'} +2026-06-28 18:23:03.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 18:23:03.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:09.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_0/index_3743_seed_21237_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 18:24:23.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:23.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:33.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_0/index_3743_seed_21238_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 18:25:53.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:25:53.541 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 18:25:54.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.725 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 414 train_steps 12560 {'step': '414/10000', 'learning_rate': '0.00001000', 'loss': '-2.701e-08', 'grad_norm': '0.00583395', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.63916016', 'grpo_loop_time': '419.31668091', 'ref_model_time': '111.47887421', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87576395', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.46093750', 'token_credit_video_advantage_mean': '-5.122e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.475e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.701e-08', 'kl_loss': '0.01045013', 'kl_moving': '0.01045013', 'kl_stepwise': '-1.971e-08', 'gathered_avg_reward_mean': '9.45983887', 'gathered_avg_reward_std': '1.12802303', 'gathered_video_score2_mq_reward_mean': '3.27709961', 'gathered_video_score2_mq_reward_std': '0.37477860', 'gathered_video_score2_overall_reward_mean': '9.46093750', 'gathered_video_score2_overall_reward_std': '1.13335121', 'gathered_video_score2_ta_reward_mean': '2.99938965', 'gathered_video_score2_ta_reward_std': '0.66238630', 'gathered_video_score2_vq_reward_mean': '3.18334961', 'gathered_video_score2_vq_reward_std': '0.36116701', 'gathered_ori_avg_reward_mean': '9.45983887', 'gathered_ori_avg_reward_std': '1.12802303'} +2026-06-28 18:36:44.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 18:36:44.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:44.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:51.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_0/index_1518_seed_21970_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 18:38:07.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:38:07.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:16.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_0/index_1518_seed_21971_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:39:38.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 18:39:39.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.160 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 415 train_steps 12600 {'step': '415/10000', 'learning_rate': '0.00001000', 'loss': '8.382e-09', 'grad_norm': '0.00498937', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.29736328', 'grpo_loop_time': '417.28198242', 'ref_model_time': '111.01931000', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86969221', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.54687500', 'token_credit_video_advantage_mean': '-8.382e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.536e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.01080890', 'kl_moving': '0.01080891', 'kl_stepwise': '-2.007e-08', 'gathered_avg_reward_mean': '9.54650879', 'gathered_avg_reward_std': '1.08796108', 'gathered_video_score2_mq_reward_mean': '3.34777832', 'gathered_video_score2_mq_reward_std': '0.41515738', 'gathered_video_score2_overall_reward_mean': '9.54687500', 'gathered_video_score2_overall_reward_std': '1.09257746', 'gathered_video_score2_ta_reward_mean': '2.97375488', 'gathered_video_score2_ta_reward_std': '0.68715364', 'gathered_video_score2_vq_reward_mean': '3.22497559', 'gathered_video_score2_vq_reward_std': '0.37524959', 'gathered_ori_avg_reward_mean': '9.54650879', 'gathered_ori_avg_reward_std': '1.08796108'} +2026-06-28 18:50:26.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 18:50:26.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:32.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_0/index_1157_seed_27008_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 18:51:49.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:49.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:00.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_0/index_1157_seed_27009_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:53:21.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 18:53:22.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.650 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 416 train_steps 12640 {'step': '416/10000', 'learning_rate': '0.00001000', 'loss': '2.664e-09', 'grad_norm': '0.01011134', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.30163574', 'grpo_loop_time': '419.13381958', 'ref_model_time': '111.22552490', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85652268', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.65576172', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.472e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '2.608e-09', 'kl_loss': '0.01221409', 'kl_moving': '0.01221409', 'kl_stepwise': '-1.984e-08', 'gathered_avg_reward_mean': '9.65869141', 'gathered_avg_reward_std': '1.02659798', 'gathered_video_score2_mq_reward_mean': '3.33300781', 'gathered_video_score2_mq_reward_std': '0.37353101', 'gathered_video_score2_overall_reward_mean': '9.65576172', 'gathered_video_score2_overall_reward_std': '1.02074265', 'gathered_video_score2_ta_reward_mean': '3.09814453', 'gathered_video_score2_ta_reward_std': '0.53298825', 'gathered_video_score2_vq_reward_mean': '3.22753906', 'gathered_video_score2_vq_reward_std': '0.38644123', 'gathered_ori_avg_reward_mean': '9.65869141', 'gathered_ori_avg_reward_std': '1.02659798'} +2026-06-28 19:04:13.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 19:04:13.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:20.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_0/index_8702_seed_7604_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 19:05:37.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:37.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:06:45.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_0/index_8702_seed_7605_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 19:07:07.109 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:07:07.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 19:07:08.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.825 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 417 train_steps 12680 {'step': '417/10000', 'learning_rate': '0.00001000', 'loss': '-1.863e-09', 'grad_norm': '0.00864408', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.86492920', 'grpo_loop_time': '420.58474731', 'ref_model_time': '111.51962280', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89506537', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.28564453', 'token_credit_video_advantage_mean': '2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.502e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.863e-09', 'kl_loss': '0.01236528', 'kl_moving': '0.01236528', 'kl_stepwise': '-1.971e-08', 'gathered_avg_reward_mean': '9.28210449', 'gathered_avg_reward_std': '1.18208539', 'gathered_video_score2_mq_reward_mean': '3.22558594', 'gathered_video_score2_mq_reward_std': '0.38932279', 'gathered_video_score2_overall_reward_mean': '9.28564453', 'gathered_video_score2_overall_reward_std': '1.18657279', 'gathered_video_score2_ta_reward_mean': '2.98083496', 'gathered_video_score2_ta_reward_std': '0.55611598', 'gathered_video_score2_vq_reward_mean': '3.07568359', 'gathered_video_score2_vq_reward_std': '0.55401981', 'gathered_ori_avg_reward_mean': '9.28210449', 'gathered_ori_avg_reward_std': '1.18208539'} +2026-06-28 19:18:01.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 19:18:01.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:01.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_0/index_3229_seed_33298_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 19:19:23.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:23.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:30.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_0/index_3229_seed_33299_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:20:52.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 19:20:53.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.278 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 418 train_steps 12720 {'step': '418/10000', 'learning_rate': '0.00001000', 'loss': '6.752e-09', 'grad_norm': '0.00582826', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.57189941', 'grpo_loop_time': '419.33862305', 'ref_model_time': '110.94867706', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86521763', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14843750', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.457e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '6.519e-09', 'kl_loss': '0.01375314', 'kl_moving': '0.01375314', 'kl_stepwise': '-2.036e-08', 'gathered_avg_reward_mean': '9.14880371', 'gathered_avg_reward_std': '0.94995934', 'gathered_video_score2_mq_reward_mean': '3.17089844', 'gathered_video_score2_mq_reward_std': '0.35203826', 'gathered_video_score2_overall_reward_mean': '9.14843750', 'gathered_video_score2_overall_reward_std': '0.94999808', 'gathered_video_score2_ta_reward_mean': '2.96972656', 'gathered_video_score2_ta_reward_std': '0.48788166', 'gathered_video_score2_vq_reward_mean': '3.00817871', 'gathered_video_score2_vq_reward_std': '0.41102827', 'gathered_ori_avg_reward_mean': '9.14880371', 'gathered_ori_avg_reward_std': '0.94995934'} +2026-06-28 19:31:44.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 19:31:44.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:50.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_0/index_7143_seed_7320_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 19:33:04.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:33:04.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:16.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_0/index_7143_seed_7321_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 19:34:36.662 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:34:36.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 19:34:37.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.724 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 419 train_steps 12760 {'step': '419/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00744902', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.73687744', 'grpo_loop_time': '417.94335938', 'ref_model_time': '110.59471130', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87701738', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.10791016', 'token_credit_video_advantage_mean': '1.024e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.485e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-10', 'kl_loss': '0.01491036', 'kl_moving': '0.01491036', 'kl_stepwise': '-1.954e-08', 'gathered_avg_reward_mean': '9.10852051', 'gathered_avg_reward_std': '1.20558381', 'gathered_video_score2_mq_reward_mean': '3.13354492', 'gathered_video_score2_mq_reward_std': '0.41441178', 'gathered_video_score2_overall_reward_mean': '9.10791016', 'gathered_video_score2_overall_reward_std': '1.20316219', 'gathered_video_score2_ta_reward_mean': '3.01562500', 'gathered_video_score2_ta_reward_std': '0.60767144', 'gathered_video_score2_vq_reward_mean': '2.95935059', 'gathered_video_score2_vq_reward_std': '0.44436958', 'gathered_ori_avg_reward_mean': '9.10852051', 'gathered_ori_avg_reward_std': '1.20558381'} +2026-06-28 19:45:24.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 19:45:24.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:24.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_0/index_2631_seed_29483_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 19:46:45.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:45.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:47:56.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_0/index_2631_seed_29484_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 19:48:17.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:48:17.980 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 19:48:18.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:07.158 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 420 train_steps 12800 {'step': '420/10000', 'learning_rate': '0.00001000', 'loss': '-8.382e-09', 'grad_norm': '0.00520570', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.43350220', 'grpo_loop_time': '418.11734009', 'ref_model_time': '111.46644592', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86522651', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.41601562', 'token_credit_video_advantage_mean': '6.054e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.443e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.211e-08', 'kl_loss': '0.01324480', 'kl_moving': '0.01324480', 'kl_stepwise': '-1.949e-08', 'gathered_avg_reward_mean': '9.41418457', 'gathered_avg_reward_std': '1.00927389', 'gathered_video_score2_mq_reward_mean': '3.21655273', 'gathered_video_score2_mq_reward_std': '0.38391569', 'gathered_video_score2_overall_reward_mean': '9.41601562', 'gathered_video_score2_overall_reward_std': '1.00392711', 'gathered_video_score2_ta_reward_mean': '3.12060547', 'gathered_video_score2_ta_reward_std': '0.66647351', 'gathered_video_score2_vq_reward_mean': '3.07702637', 'gathered_video_score2_vq_reward_std': '0.39138079', 'gathered_ori_avg_reward_mean': '9.41418457', 'gathered_ori_avg_reward_std': '1.00927389'} +2026-06-28 19:59:42.644 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 19:59:44.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:50.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_0/index_5321_seed_8604_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 20:01:04.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:01:04.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:15.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_0/index_5321_seed_8605_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:02:34.519 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 20:02:35.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.749 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 421 train_steps 12840 {'step': '421/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00553231', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '282.74578857', 'grpo_loop_time': '419.21896362', 'ref_model_time': '110.94748688', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88170946', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.34765625', 'token_credit_video_advantage_mean': '-9.313e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.429e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.00937039', 'kl_moving': '0.00937039', 'kl_stepwise': '-1.984e-08', 'gathered_avg_reward_mean': '9.35180664', 'gathered_avg_reward_std': '1.41922522', 'gathered_video_score2_mq_reward_mean': '3.17883301', 'gathered_video_score2_mq_reward_std': '0.45129681', 'gathered_video_score2_overall_reward_mean': '9.34765625', 'gathered_video_score2_overall_reward_std': '1.42006564', 'gathered_video_score2_ta_reward_mean': '3.15405273', 'gathered_video_score2_ta_reward_std': '0.68978155', 'gathered_video_score2_vq_reward_mean': '3.01892090', 'gathered_video_score2_vq_reward_std': '0.51718146', 'gathered_ori_avg_reward_mean': '9.35180664', 'gathered_ori_avg_reward_std': '1.41922522'} +2026-06-28 20:13:24.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 20:13:24.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:24.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:31.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_0/index_7063_seed_14591_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 20:14:47.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:47.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:15:56.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_0/index_7063_seed_14592_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:16:15.209 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 20:16:16.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.078 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 422 train_steps 12880 {'step': '422/10000', 'learning_rate': '0.00001000', 'loss': '-6.519e-09', 'grad_norm': '0.00687495', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.55413818', 'grpo_loop_time': '419.75634766', 'ref_model_time': '111.50237274', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88498056', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.67431641', 'token_credit_video_advantage_mean': '1.118e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.527e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.00980195', 'kl_moving': '0.00980196', 'kl_stepwise': '-1.949e-08', 'gathered_avg_reward_mean': '9.67761230', 'gathered_avg_reward_std': '1.01778173', 'gathered_video_score2_mq_reward_mean': '3.28149414', 'gathered_video_score2_mq_reward_std': '0.41266432', 'gathered_video_score2_overall_reward_mean': '9.67431641', 'gathered_video_score2_overall_reward_std': '1.01743841', 'gathered_video_score2_ta_reward_mean': '3.26159668', 'gathered_video_score2_ta_reward_std': '0.47307572', 'gathered_video_score2_vq_reward_mean': '3.13452148', 'gathered_video_score2_vq_reward_std': '0.46322110', 'gathered_ori_avg_reward_mean': '9.67761230', 'gathered_ori_avg_reward_std': '1.01778173'} +2026-06-28 20:27:07.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 20:27:07.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:13.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_0/index_7252_seed_27272_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 20:28:33.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:33.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:38.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_0/index_7252_seed_27273_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:29:58.591 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 20:29:59.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.245 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.909 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 423 train_steps 12920 {'step': '423/10000', 'learning_rate': '0.00001000', 'loss': '1.863e-08', 'grad_norm': '0.00525099', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.58465576', 'grpo_loop_time': '418.14306641', 'ref_model_time': '112.43608093', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86484194', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.29589844', 'token_credit_video_advantage_mean': '5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.459e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-08', 'kl_loss': '0.00986449', 'kl_moving': '0.00986449', 'kl_stepwise': '-2.009e-08', 'gathered_avg_reward_mean': '9.29711914', 'gathered_avg_reward_std': '1.29409754', 'gathered_video_score2_mq_reward_mean': '3.17517090', 'gathered_video_score2_mq_reward_std': '0.43768919', 'gathered_video_score2_overall_reward_mean': '9.29589844', 'gathered_video_score2_overall_reward_std': '1.29259658', 'gathered_video_score2_ta_reward_mean': '2.95800781', 'gathered_video_score2_ta_reward_std': '0.68280715', 'gathered_video_score2_vq_reward_mean': '3.16394043', 'gathered_video_score2_vq_reward_std': '0.40354013', 'gathered_ori_avg_reward_mean': '9.29711914', 'gathered_ori_avg_reward_std': '1.29409754'} +2026-06-28 20:40:48.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 20:40:49.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_0/index_6425_seed_15667_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 20:42:17.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:42:17.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:23.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_0/index_6425_seed_15668_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:43:45.493 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 20:43:46.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.443 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 424 train_steps 12960 {'step': '424/10000', 'learning_rate': '0.00001000', 'loss': '1.770e-08', 'grad_norm': '0.00583182', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '289.25631714', 'grpo_loop_time': '418.57269287', 'ref_model_time': '110.94616699', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.89939737', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.43896484', 'token_credit_video_advantage_mean': '-3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.455e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.770e-08', 'kl_loss': '0.00994975', 'kl_moving': '0.00994976', 'kl_stepwise': '-1.932e-08', 'gathered_avg_reward_mean': '9.44189453', 'gathered_avg_reward_std': '1.19140291', 'gathered_video_score2_mq_reward_mean': '3.23876953', 'gathered_video_score2_mq_reward_std': '0.40329519', 'gathered_video_score2_overall_reward_mean': '9.43896484', 'gathered_video_score2_overall_reward_std': '1.18882823', 'gathered_video_score2_ta_reward_mean': '3.01330566', 'gathered_video_score2_ta_reward_std': '0.59725827', 'gathered_video_score2_vq_reward_mean': '3.18981934', 'gathered_video_score2_vq_reward_std': '0.44131482', 'gathered_ori_avg_reward_mean': '9.44189453', 'gathered_ori_avg_reward_std': '1.19140291'} +2026-06-28 20:54:36.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 20:54:36.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:42.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_0/index_5541_seed_27371_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 20:55:58.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:58.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:09.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_0/index_5541_seed_27372_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:57:33.609 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 20:57:34.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:08:24.132 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 425 train_steps 13000 {'step': '425/10000', 'learning_rate': '0.00001000', 'loss': '-6.519e-09', 'grad_norm': '0.00587518', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '290.03775024', 'grpo_loop_time': '419.82659912', 'ref_model_time': '111.24324799', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85932118', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.41503906', 'token_credit_video_advantage_mean': '-4.424e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.444e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-8.382e-09', 'kl_loss': '0.01109385', 'kl_moving': '0.01109385', 'kl_stepwise': '-1.930e-08', 'gathered_avg_reward_mean': '9.41320801', 'gathered_avg_reward_std': '1.36530578', 'gathered_video_score2_mq_reward_mean': '3.25378418', 'gathered_video_score2_mq_reward_std': '0.44460306', 'gathered_video_score2_overall_reward_mean': '9.41503906', 'gathered_video_score2_overall_reward_std': '1.36428785', 'gathered_video_score2_ta_reward_mean': '2.98999023', 'gathered_video_score2_ta_reward_std': '0.73858404', 'gathered_video_score2_vq_reward_mean': '3.16943359', 'gathered_video_score2_vq_reward_std': '0.39813310', 'gathered_ori_avg_reward_mean': '9.41320801', 'gathered_ori_avg_reward_std': '1.36530566'} +2026-06-28 21:09:41.136 | INFO | __main__:save_checkpoint:1967 - Timer checkpoint saved at step 425 to s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-425 +2026-06-28 21:09:41.271 | INFO | __main__:_write_timer_marker:1784 - Timer resume marker updated at step 425: s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer/checkpoint-425 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 21:09:41.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_0/index_1153_seed_29891_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 21:11:07.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:11:07.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:14.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_0/index_1153_seed_29892_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:12:34.807 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 21:12:35.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.733 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 426 train_steps 13040 {'step': '426/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00666090', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.94384766', 'grpo_loop_time': '418.42092896', 'ref_model_time': '110.85316467', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86343586', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.31298828', 'token_credit_video_advantage_mean': '-6.519e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.415e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.01140592', 'kl_moving': '0.01140592', 'kl_stepwise': '-1.893e-08', 'gathered_avg_reward_mean': '9.31774902', 'gathered_avg_reward_std': '1.44856918', 'gathered_video_score2_mq_reward_mean': '3.17138672', 'gathered_video_score2_mq_reward_std': '0.44097149', 'gathered_video_score2_overall_reward_mean': '9.31298828', 'gathered_video_score2_overall_reward_std': '1.45029223', 'gathered_video_score2_ta_reward_mean': '3.12500000', 'gathered_video_score2_ta_reward_std': '0.69258595', 'gathered_video_score2_vq_reward_mean': '3.02136230', 'gathered_video_score2_vq_reward_std': '0.55242586', 'gathered_ori_avg_reward_mean': '9.31774902', 'gathered_ori_avg_reward_std': '1.44856918'} +2026-06-28 21:23:23.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 21:23:23.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:23.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_0/index_7797_seed_16598_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 21:24:43.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:43.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:25:53.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_0/index_7797_seed_16599_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 21:26:14.845 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:26:15.013 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 21:26:15.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.272 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 427 train_steps 13080 {'step': '427/10000', 'learning_rate': '0.00001000', 'loss': '1.723e-08', 'grad_norm': '0.00574387', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.53021240', 'grpo_loop_time': '420.02615356', 'ref_model_time': '111.68613434', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87439966', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.09375000', 'token_credit_video_advantage_mean': '-1.863e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.407e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.723e-08', 'kl_loss': '0.01173666', 'kl_moving': '0.01173666', 'kl_stepwise': '-1.940e-08', 'gathered_avg_reward_mean': '9.09936523', 'gathered_avg_reward_std': '0.93438756', 'gathered_video_score2_mq_reward_mean': '3.17993164', 'gathered_video_score2_mq_reward_std': '0.33715025', 'gathered_video_score2_overall_reward_mean': '9.09375000', 'gathered_video_score2_overall_reward_std': '0.93276799', 'gathered_video_score2_ta_reward_mean': '2.93530273', 'gathered_video_score2_ta_reward_std': '0.46395552', 'gathered_video_score2_vq_reward_mean': '2.98413086', 'gathered_video_score2_vq_reward_std': '0.44767475', 'gathered_ori_avg_reward_mean': '9.09936523', 'gathered_ori_avg_reward_std': '0.93438762'} +2026-06-28 21:37:07.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 21:37:07.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:13.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_0/index_254_seed_20544_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 21:38:28.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:28.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:36.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_0/index_254_seed_20545_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:39:57.778 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 21:39:58.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.556 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 428 train_steps 13120 {'step': '428/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00534319', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.66796875', 'grpo_loop_time': '418.21624756', 'ref_model_time': '112.61325073', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85926807', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.14355469', 'token_credit_video_advantage_mean': '7.916e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.413e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '4.657e-09', 'kl_loss': '0.01253126', 'kl_moving': '0.01253126', 'kl_stepwise': '-1.866e-08', 'gathered_avg_reward_mean': '9.14611816', 'gathered_avg_reward_std': '0.95928329', 'gathered_video_score2_mq_reward_mean': '3.20776367', 'gathered_video_score2_mq_reward_std': '0.32045156', 'gathered_video_score2_overall_reward_mean': '9.14355469', 'gathered_video_score2_overall_reward_std': '0.95991623', 'gathered_video_score2_ta_reward_mean': '2.85278320', 'gathered_video_score2_ta_reward_std': '0.64460409', 'gathered_video_score2_vq_reward_mean': '3.08557129', 'gathered_video_score2_vq_reward_std': '0.39624241', 'gathered_ori_avg_reward_mean': '9.14611816', 'gathered_ori_avg_reward_std': '0.95928335'} +2026-06-28 21:50:48.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 21:50:48.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:55.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_0/index_8879_seed_15331_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 21:52:07.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:52:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_0/index_8879_seed_15332_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:53:42.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 21:53:43.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.660 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 429 train_steps 13160 {'step': '429/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-10', 'grad_norm': '0.00579495', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.69464111', 'grpo_loop_time': '417.73339844', 'ref_model_time': '111.19404602', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87069499', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.23388672', 'token_credit_video_advantage_mean': '0.00000000', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.454e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.794e-09', 'kl_loss': '0.01334394', 'kl_moving': '0.01334394', 'kl_stepwise': '-1.971e-08', 'gathered_avg_reward_mean': '9.23193359', 'gathered_avg_reward_std': '1.11303782', 'gathered_video_score2_mq_reward_mean': '3.22827148', 'gathered_video_score2_mq_reward_std': '0.35727367', 'gathered_video_score2_overall_reward_mean': '9.23388672', 'gathered_video_score2_overall_reward_std': '1.11278081', 'gathered_video_score2_ta_reward_mean': '2.85278320', 'gathered_video_score2_ta_reward_std': '0.72674125', 'gathered_video_score2_vq_reward_mean': '3.15087891', 'gathered_video_score2_vq_reward_std': '0.31202623', 'gathered_ori_avg_reward_mean': '9.23193359', 'gathered_ori_avg_reward_std': '1.11303782'} +2026-06-28 22:04:31.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 22:04:31.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:31.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_0/index_1973_seed_8341_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 22:05:52.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:52.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:05.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_0/index_1973_seed_8342_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:07:25.912 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 22:07:26.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:15.448 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 430 train_steps 13200 {'step': '430/10000', 'learning_rate': '0.00001000', 'loss': '-1.192e-08', 'grad_norm': '0.00537833', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.14312744', 'grpo_loop_time': '417.23010254', 'ref_model_time': '111.35631561', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.87007928', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.24365234', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.478e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-1.229e-08', 'kl_loss': '0.01376910', 'kl_moving': '0.01376910', 'kl_stepwise': '-2.078e-08', 'gathered_avg_reward_mean': '9.24450684', 'gathered_avg_reward_std': '1.21452868', 'gathered_video_score2_mq_reward_mean': '3.23950195', 'gathered_video_score2_mq_reward_std': '0.39830396', 'gathered_video_score2_overall_reward_mean': '9.24365234', 'gathered_video_score2_overall_reward_std': '1.21546757', 'gathered_video_score2_ta_reward_mean': '2.96166992', 'gathered_video_score2_ta_reward_std': '0.62093973', 'gathered_video_score2_vq_reward_mean': '3.04333496', 'gathered_video_score2_vq_reward_std': '0.43419144', 'gathered_ori_avg_reward_mean': '9.24450684', 'gathered_ori_avg_reward_std': '1.21452868'} +2026-06-28 22:18:50.804 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 22:18:53.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:59.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_0/index_4950_seed_7551_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 22:20:12.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:20:12.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:24.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_0/index_4950_seed_7552_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 22:21:46.144 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:21:46.297 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 22:21:47.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.097 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 431 train_steps 13240 {'step': '431/10000', 'learning_rate': '0.00001000', 'loss': '1.304e-09', 'grad_norm': '0.00586524', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.98046875', 'grpo_loop_time': '420.37484741', 'ref_model_time': '111.72910309', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.91371566', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.11718750', 'token_credit_video_advantage_mean': '-2.328e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.335e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-2.421e-09', 'kl_loss': '0.00942563', 'kl_moving': '0.00942564', 'kl_stepwise': '-1.836e-08', 'gathered_avg_reward_mean': '9.11767578', 'gathered_avg_reward_std': '1.10562909', 'gathered_video_score2_mq_reward_mean': '3.12878418', 'gathered_video_score2_mq_reward_std': '0.39669356', 'gathered_video_score2_overall_reward_mean': '9.11718750', 'gathered_video_score2_overall_reward_std': '1.10392869', 'gathered_video_score2_ta_reward_mean': '2.92370605', 'gathered_video_score2_ta_reward_std': '0.64434588', 'gathered_video_score2_vq_reward_mean': '3.06518555', 'gathered_video_score2_vq_reward_std': '0.35084078', 'gathered_ori_avg_reward_mean': '9.11767578', 'gathered_ori_avg_reward_std': '1.10562909'} +2026-06-28 22:32:40.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 22:32:40.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:46.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_0/index_339_seed_6790_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 22:34:05.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:34:05.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:12.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_0/index_339_seed_6791_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:35:34.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 22:35:35.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:23.991 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 432 train_steps 13280 {'step': '432/10000', 'learning_rate': '0.00001000', 'loss': '4.657e-09', 'grad_norm': '0.00491121', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.65652466', 'grpo_loop_time': '418.18261719', 'ref_model_time': '111.72078705', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84920073', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.38916016', 'token_credit_video_advantage_mean': '-1.397e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.457e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '3.725e-09', 'kl_loss': '0.00889506', 'kl_moving': '0.00889506', 'kl_stepwise': '-1.939e-08', 'gathered_avg_reward_mean': '9.39001465', 'gathered_avg_reward_std': '0.92771107', 'gathered_video_score2_mq_reward_mean': '3.31347656', 'gathered_video_score2_mq_reward_std': '0.35620710', 'gathered_video_score2_overall_reward_mean': '9.38916016', 'gathered_video_score2_overall_reward_std': '0.92501336', 'gathered_video_score2_ta_reward_mean': '2.76110840', 'gathered_video_score2_ta_reward_std': '0.73467988', 'gathered_video_score2_vq_reward_mean': '3.31542969', 'gathered_video_score2_vq_reward_std': '0.27502158', 'gathered_ori_avg_reward_mean': '9.39001465', 'gathered_ori_avg_reward_std': '0.92771107'} +2026-06-28 22:46:24.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 22:46:24.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:30.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_0/index_9724_seed_7006_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 22:47:45.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:45.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:48:57.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_0/index_9724_seed_7007_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:49:17.695 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 22:49:18.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:06.946 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 433 train_steps 13320 {'step': '433/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00625391', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '287.22552490', 'grpo_loop_time': '417.81564331', 'ref_model_time': '111.73587799', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85081196', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.45068359', 'token_credit_video_advantage_mean': '-5.588e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.462e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '7.451e-09', 'kl_loss': '0.01032798', 'kl_moving': '0.01032799', 'kl_stepwise': '-1.972e-08', 'gathered_avg_reward_mean': '9.44592285', 'gathered_avg_reward_std': '1.15335238', 'gathered_video_score2_mq_reward_mean': '3.24389648', 'gathered_video_score2_mq_reward_std': '0.36108646', 'gathered_video_score2_overall_reward_mean': '9.45068359', 'gathered_video_score2_overall_reward_std': '1.15029347', 'gathered_video_score2_ta_reward_mean': '3.09973145', 'gathered_video_score2_ta_reward_std': '0.57535982', 'gathered_video_score2_vq_reward_mean': '3.10229492', 'gathered_video_score2_vq_reward_std': '0.42891380', 'gathered_ori_avg_reward_mean': '9.44592285', 'gathered_ori_avg_reward_std': '1.15335238'} +2026-06-28 23:00:06.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 23:00:07.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:13.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_0/index_1210_seed_31083_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 23:01:27.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:27.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:38.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_0/index_1210_seed_31084_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 23:02:58.268 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:02:58.408 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 23:02:59.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.692 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 434 train_steps 13360 {'step': '434/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00539307', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.73474121', 'grpo_loop_time': '418.08859253', 'ref_model_time': '113.64681244', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85331362', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.75732422', 'token_credit_video_advantage_mean': '3.725e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.456e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.00989551', 'kl_moving': '0.00989551', 'kl_stepwise': '-2.000e-08', 'gathered_avg_reward_mean': '9.75915527', 'gathered_avg_reward_std': '1.13149893', 'gathered_video_score2_mq_reward_mean': '3.32348633', 'gathered_video_score2_mq_reward_std': '0.42456713', 'gathered_video_score2_overall_reward_mean': '9.75732422', 'gathered_video_score2_overall_reward_std': '1.13283598', 'gathered_video_score2_ta_reward_mean': '3.22644043', 'gathered_video_score2_ta_reward_std': '0.58186001', 'gathered_video_score2_vq_reward_mean': '3.20922852', 'gathered_video_score2_vq_reward_std': '0.43278021', 'gathered_ori_avg_reward_mean': '9.75915527', 'gathered_ori_avg_reward_std': '1.13149893'} +2026-06-28 23:13:49.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 23:13:49.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:55.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_0/index_3071_seed_949_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 23:15:11.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:15:11.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:20.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_0/index_3071_seed_950_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:16:43.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 23:16:44.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.786 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 435 train_steps 13400 {'step': '435/10000', 'learning_rate': '0.00001000', 'loss': '-5.588e-09', 'grad_norm': '0.00617331', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.88183594', 'grpo_loop_time': '417.62527466', 'ref_model_time': '111.35119629', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86192852', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.61523438', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.459e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '-5.588e-09', 'kl_loss': '0.01047607', 'kl_moving': '0.01047607', 'kl_stepwise': '-1.912e-08', 'gathered_avg_reward_mean': '9.60876465', 'gathered_avg_reward_std': '1.15279746', 'gathered_video_score2_mq_reward_mean': '3.27197266', 'gathered_video_score2_mq_reward_std': '0.41111976', 'gathered_video_score2_overall_reward_mean': '9.61523438', 'gathered_video_score2_overall_reward_std': '1.15536380', 'gathered_video_score2_ta_reward_mean': '3.22839355', 'gathered_video_score2_ta_reward_std': '0.50172079', 'gathered_video_score2_vq_reward_mean': '3.10839844', 'gathered_video_score2_vq_reward_std': '0.42884216', 'gathered_ori_avg_reward_mean': '9.60876465', 'gathered_ori_avg_reward_std': '1.15279746'} +2026-06-28 23:27:31.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 23:27:32.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:38.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_0/index_2146_seed_6499_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 23:28:52.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:52.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:03.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_0/index_2146_seed_6500_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:30:22.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 23:30:23.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:13.936 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 436 train_steps 13440 {'step': '436/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00472213', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '283.24810791', 'grpo_loop_time': '418.67822266', 'ref_model_time': '111.19300079', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84820902', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.06494141', 'token_credit_video_advantage_mean': '-3.260e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.439e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.211e-08', 'kl_loss': '0.01067142', 'kl_moving': '0.01067143', 'kl_stepwise': '-1.893e-08', 'gathered_avg_reward_mean': '9.06311035', 'gathered_avg_reward_std': '1.32454586', 'gathered_video_score2_mq_reward_mean': '3.17956543', 'gathered_video_score2_mq_reward_std': '0.41538128', 'gathered_video_score2_overall_reward_mean': '9.06494141', 'gathered_video_score2_overall_reward_std': '1.32329440', 'gathered_video_score2_ta_reward_mean': '2.89904785', 'gathered_video_score2_ta_reward_std': '0.63881719', 'gathered_video_score2_vq_reward_mean': '2.98449707', 'gathered_video_score2_vq_reward_std': '0.53741205', 'gathered_ori_avg_reward_mean': '9.06311035', 'gathered_ori_avg_reward_std': '1.32454586'} +2026-06-28 23:41:14.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 23:41:14.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:20.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_0/index_4267_seed_15323_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 23:42:35.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:35.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:43:45.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_0/index_4267_seed_15324_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 23:44:05.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:44:05.724 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 23:44:06.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.041 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 437 train_steps 13480 {'step': '437/10000', 'learning_rate': '0.00001000', 'loss': '7.451e-09', 'grad_norm': '0.00864381', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '284.52154541', 'grpo_loop_time': '421.46066284', 'ref_model_time': '111.13854980', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88234961', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.56103516', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.410e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.863e-09', 'kl_loss': '0.01087322', 'kl_moving': '0.01087322', 'kl_stepwise': '-1.814e-08', 'gathered_avg_reward_mean': '9.56518555', 'gathered_avg_reward_std': '1.33776188', 'gathered_video_score2_mq_reward_mean': '3.28771973', 'gathered_video_score2_mq_reward_std': '0.46420023', 'gathered_video_score2_overall_reward_mean': '9.56103516', 'gathered_video_score2_overall_reward_std': '1.33889735', 'gathered_video_score2_ta_reward_mean': '3.04833984', 'gathered_video_score2_ta_reward_std': '0.72560745', 'gathered_video_score2_vq_reward_mean': '3.22912598', 'gathered_video_score2_vq_reward_std': '0.41326034', 'gathered_ori_avg_reward_mean': '9.56518555', 'gathered_ori_avg_reward_std': '1.33776176'} +2026-06-28 23:55:00.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-28 23:55:00.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:06.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_0/index_6133_seed_15125_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-28 23:56:19.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:19.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:29.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_0/index_6133_seed_15126_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:57:49.084 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-28 23:57:50.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.240 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 438 train_steps 13520 {'step': '438/10000', 'learning_rate': '0.00001000', 'loss': '9.313e-09', 'grad_norm': '0.00613951', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '281.67242432', 'grpo_loop_time': '419.16616821', 'ref_model_time': '111.00798035', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.84124690', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.21777344', 'token_credit_video_advantage_mean': '-2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '9.313e-09', 'kl_loss': '0.01197265', 'kl_moving': '0.01197266', 'kl_stepwise': '-1.919e-08', 'gathered_avg_reward_mean': '9.21643066', 'gathered_avg_reward_std': '0.88721514', 'gathered_video_score2_mq_reward_mean': '3.16137695', 'gathered_video_score2_mq_reward_std': '0.24609460', 'gathered_video_score2_overall_reward_mean': '9.21777344', 'gathered_video_score2_overall_reward_std': '0.88932145', 'gathered_video_score2_ta_reward_mean': '2.99938965', 'gathered_video_score2_ta_reward_std': '0.52510434', 'gathered_video_score2_vq_reward_mean': '3.05566406', 'gathered_video_score2_vq_reward_std': '0.38709217', 'gathered_ori_avg_reward_mean': '9.21643066', 'gathered_ori_avg_reward_std': '0.88721514'} +2026-06-29 00:08:39.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-29 00:08:39.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:45.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_0/index_8474_seed_12862_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-29 00:10:00.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:10:00.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:08.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_0/index_8474_seed_12863_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:11:28.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-29 00:11:29.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.198 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 439 train_steps 13560 {'step': '439/10000', 'learning_rate': '0.00001000', 'loss': '5.588e-09', 'grad_norm': '0.00645845', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '281.53320312', 'grpo_loop_time': '419.23941040', 'ref_model_time': '110.88667297', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.86598623', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.74169922', 'token_credit_video_advantage_mean': '-4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.439e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '5.588e-09', 'kl_loss': '0.01122006', 'kl_moving': '0.01122006', 'kl_stepwise': '-1.920e-08', 'gathered_avg_reward_mean': '8.74121094', 'gathered_avg_reward_std': '1.16370809', 'gathered_video_score2_mq_reward_mean': '3.14099121', 'gathered_video_score2_mq_reward_std': '0.42072603', 'gathered_video_score2_overall_reward_mean': '8.74169922', 'gathered_video_score2_overall_reward_std': '1.15936339', 'gathered_video_score2_ta_reward_mean': '2.60046387', 'gathered_video_score2_ta_reward_std': '0.55557203', 'gathered_video_score2_vq_reward_mean': '2.99975586', 'gathered_video_score2_vq_reward_std': '0.53764880', 'gathered_ori_avg_reward_mean': '8.74121094', 'gathered_ori_avg_reward_std': '1.16370809'} +2026-06-29 00:22:19.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-29 00:22:19.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:25.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_0/index_7619_seed_29339_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-29 00:23:40.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:40.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:24:52.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_0/index_7619_seed_29340_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:25:12.669 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-29 00:25:13.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:05.902 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 440 train_steps 13600 {'step': '440/10000', 'learning_rate': '0.00001000', 'loss': '1.183e-08', 'grad_norm': '0.00569829', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '286.71215820', 'grpo_loop_time': '420.21142578', 'ref_model_time': '111.70608521', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.88351941', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '8.90673828', 'token_credit_video_advantage_mean': '-1.490e-08', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.424e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '1.183e-08', 'kl_loss': '0.01324333', 'kl_moving': '0.01324333', 'kl_stepwise': '-1.959e-08', 'gathered_avg_reward_mean': '8.90759277', 'gathered_avg_reward_std': '1.08464885', 'gathered_video_score2_mq_reward_mean': '3.08056641', 'gathered_video_score2_mq_reward_std': '0.39139786', 'gathered_video_score2_overall_reward_mean': '8.90673828', 'gathered_video_score2_overall_reward_std': '1.08155155', 'gathered_video_score2_ta_reward_mean': '2.89501953', 'gathered_video_score2_ta_reward_std': '0.62938201', 'gathered_video_score2_vq_reward_mean': '2.93200684', 'gathered_video_score2_vq_reward_std': '0.46787167', 'gathered_ori_avg_reward_mean': '8.90759277', 'gathered_ori_avg_reward_std': '1.08464885'} +2026-06-29 00:36:41.444 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-29 00:36:43.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_0/index_2207_seed_3092_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-29 00:38:10.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:38:10.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:15.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_0/index_2207_seed_3093_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-29 00:39:38.559 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:39:38.722 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-29 00:39:39.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.053 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 441 train_steps 13640 {'step': '441/10000', 'learning_rate': '0.00001000', 'loss': '1.211e-08', 'grad_norm': '0.00524421', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '288.17639160', 'grpo_loop_time': '417.96682739', 'ref_model_time': '110.78804779', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85266852', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.01123047', 'token_credit_video_advantage_mean': '4.657e-10', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.476e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '8.382e-09', 'kl_loss': '0.00932122', 'kl_moving': '0.00932122', 'kl_stepwise': '-1.977e-08', 'gathered_avg_reward_mean': '9.01220703', 'gathered_avg_reward_std': '1.17079282', 'gathered_video_score2_mq_reward_mean': '3.15112305', 'gathered_video_score2_mq_reward_std': '0.41487294', 'gathered_video_score2_overall_reward_mean': '9.01123047', 'gathered_video_score2_overall_reward_std': '1.17236149', 'gathered_video_score2_ta_reward_mean': '2.88500977', 'gathered_video_score2_ta_reward_std': '0.67154914', 'gathered_video_score2_vq_reward_mean': '2.97607422', 'gathered_video_score2_vq_reward_std': '0.42222497', 'gathered_ori_avg_reward_mean': '9.01220703', 'gathered_ori_avg_reward_std': '1.17079282'} +2026-06-29 00:50:27.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-29 00:50:27.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:33.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_0/index_8113_seed_19908_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-29 00:51:47.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:47.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:52:57.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_0/index_8113_seed_19909_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-29 00:53:18.756 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:53:18.920 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-29 00:53:19.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.326 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 0, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.084 | INFO | __main__:_log_progress_step:2147 - Training Progress: update_steps 442 train_steps 13680 {'step': '442/10000', 'learning_rate': '0.00001000', 'loss': '0.00000000', 'grad_norm': '0.00568730', 'KL_weight': '1.000e-07', 'param_delta_mean': '0.00000000', 'param_delta_max': '0.00000000', 'rollout_time': '285.58245850', 'grpo_loop_time': '418.10937500', 'ref_model_time': '112.15598297', 'token_credit_ratio_mean': '1.00000000', 'token_credit_advantage_abs_mean': '0.85420132', 'token_credit_question_count': '1.00000000', 'token_credit_video_logprob_reward_mean': '9.43652344', 'token_credit_video_advantage_mean': '2.794e-09', 'token_credit_approx_kl': '0.00000000', 'approx_kl': '1.483e-15', 'clipfrac': '0.00000000', 'clipfrac_gt_one': '0.00000000', 'clipfrac_lt_one': '0.00000000', 'policy_loss': '0.00000000', 'kl_loss': '0.00873519', 'kl_moving': '0.00873520', 'kl_stepwise': '-1.844e-08', 'gathered_avg_reward_mean': '9.43566895', 'gathered_avg_reward_std': '1.15201497', 'gathered_video_score2_mq_reward_mean': '3.24865723', 'gathered_video_score2_mq_reward_std': '0.40698540', 'gathered_video_score2_overall_reward_mean': '9.43652344', 'gathered_video_score2_overall_reward_std': '1.15647388', 'gathered_video_score2_ta_reward_mean': '2.94982910', 'gathered_video_score2_ta_reward_std': '0.73217160', 'gathered_video_score2_vq_reward_mean': '3.23718262', 'gathered_video_score2_vq_reward_std': '0.40206483', 'gathered_ori_avg_reward_mean': '9.43566895', 'gathered_ori_avg_reward_std': '1.15201497'} +2026-06-29 01:04:09.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-06-29 01:04:09.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:16.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_0/index_2904_seed_5954_dp_0_rank_0_sample_0_slot_0.mp4 +2026-06-29 01:05:30.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:30.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:06:41.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_0/index_2904_seed_5955_dp_0_rank_0_sample_1_slot_1.mp4 +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 01:07:02.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-06-29 01:07:04.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_19.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_19.log new file mode 100644 index 0000000000000000000000000000000000000000..f9c15fb0090b9e9e023f5ea4fc77eeae7a904007 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_19.log @@ -0,0 +1,5147 @@ +2026-05-29 20:45:52.442 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:45:52.445 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 20:51:46.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:10.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:10.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 19, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 21:06:57.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:01.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 21:21:17.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:15.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 21:35:35.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:43.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 21:49:56.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:03.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:04.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.129 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 22:04:18.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 22:18:35.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:41.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 22:32:58.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:00.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.542 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 22:47:20.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:28.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 23:01:44.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:51.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:51.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:49.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:52.825 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 23:16:57.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:58.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 23:31:14.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 23:45:29.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:35.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:35.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-29 23:59:50.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:57.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 00:14:13.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 00:28:36.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:44.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:44.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 00:42:58.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:06.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 00:57:15.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:21.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:21.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 01:11:38.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 01:26:05.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.678 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 01:41:19.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:28.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:18.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 01:55:38.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:44.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:39.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 02:09:58.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:02.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 02:24:23.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:30.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:30.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 02:38:56.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:02.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:02.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:57.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 02:54:31.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:37.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:37.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 03:08:50.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:56.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:56.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:53.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 03:23:09.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:16.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:16.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.442 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 03:37:26.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 03:51:59.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:06.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.615 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 04:07:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 04:21:48.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:57.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:57.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 04:36:12.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 04:50:41.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 05:05:07.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:12.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:12.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:12.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.655 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 05:19:25.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 05:33:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:44.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:44.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:37.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.262 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 05:47:59.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:58.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 06:02:21.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 06:16:54.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 06:19:59.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:12.048 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 06:32:17.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:16.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 06:46:37.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:42.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:42.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:38.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 07:00:59.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:07.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:00.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 07:15:21.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:22.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:41.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 07:30:08.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.021 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 07:44:30.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:37.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:37.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 07:59:00.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:08.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 08:13:33.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:40.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:40.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 08:28:15.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:23.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:23.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:23.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 08:42:53.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:01.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:01.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:25.025 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 08:59:29.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:38.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:38.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:33.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 09:14:04.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:11.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:16.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.550 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 09:28:42.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:43.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:50.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:50.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 09:43:17.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:24.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 09:57:46.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:53.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 10:12:07.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:06.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 10:26:24.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:30.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:30.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.324 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 10:40:46.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:53.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:53.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 10:55:08.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:09.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:16.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:16.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 11:09:40.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:48.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:48.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:03.018 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 11:25:07.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:16.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:16.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.076 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.464 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 11:39:42.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:49.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:49.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 11:54:06.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 12:08:30.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:36.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:36.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:30.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 12:22:53.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:01.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:01.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 12:37:12.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:18.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.052 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:11.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 12:51:30.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 13:05:59.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:06.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:04.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 13:20:24.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:31.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:31.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:30.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 13:34:48.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.289 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 13:50:02.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:10.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:10.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 14:04:33.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 14:19:43.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:54.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:54.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 14:34:38.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 14:49:20.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:28.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:28.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 15:05:01.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:09.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:03.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 15:19:35.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 15:34:11.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:20.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:20.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 15:48:48.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:50.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.484 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 16:03:17.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:25.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.453 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 16:18:48.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:01.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 16:33:33.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 16:48:15.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 17:02:55.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 17:17:35.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:43.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:43.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 17:32:10.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 17:46:54.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:02.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:02.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 18:01:35.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 18:16:18.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 18:31:18.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:17.766 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 18:47:24.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:39.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 19:02:13.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:34.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 19:17:12.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:26.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 19:31:55.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 19:47:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:11.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 20:01:46.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 20:16:29.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 20:31:20.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 20:46:03.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:11.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:11.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 21:00:47.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.143 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 21:17:31.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:40.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:40.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:37.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 21:32:08.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:48:42.709 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:48:42.709 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 21:55:22.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:47.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:47.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:07.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 19, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 22:10:52.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:11:59.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 22:25:33.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:41.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:41.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 22:40:03.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:11.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:11.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:10.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 22:54:38.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:46.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:50.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 23:09:20.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:28.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 23:24:05.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:14.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 23:38:46.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:54.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-30 23:53:10.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:17.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:17.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:15.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 00:07:41.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:53.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:54.165 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 00:22:59.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 00:37:30.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:39.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 00:52:07.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:13.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:13.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:22.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 01:06:44.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:52.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:52.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 01:21:16.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:17.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 01:35:36.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:43.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 01:50:03.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:10.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:06.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.331 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 02:04:34.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:41.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:41.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.764 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:39.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 19, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-05-31 02:19:01.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:07.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:07.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:02.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:00:43.162 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:00:43.165 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 18:04:09.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:30.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:30.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 19, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 18:18:41.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:35.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 18:32:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 18:46:08.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:13.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:03.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 19:00:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 19:13:49.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 19:27:32.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:24.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 19:41:16.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:22.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 19:55:08.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:14.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 20:09:02.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.727 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.891 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 20:23:32.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 20:37:18.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:19.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:24.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 20:51:03.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:56.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 21:04:50.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:56.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 21:18:43.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:49.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:49.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 21:32:31.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 21:46:16.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 22:00:05.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:11.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 22:13:50.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:55.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:55.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 22:27:39.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:44.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:57.796 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 22:42:00.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 22:55:45.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:50.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 23:09:30.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:35.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 23:23:15.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:20.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:11.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 23:37:07.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:12.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:02.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-25 23:51:52.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 00:05:36.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 00:19:21.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:26.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 00:33:07.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 00:47:03.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:00.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:29.125 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 01:01:30.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 01:15:17.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:23.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:12.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 01:29:08.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:13.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:13.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.423 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 01:42:58.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 01:56:44.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 02:10:31.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 02:24:16.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 02:38:08.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:13.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 02:41:00.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 02:51:53.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:59.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 03:05:40.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:45.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:46.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.774 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 03:20:06.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 03:33:52.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:57.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:57.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 03:47:40.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:45.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 04:01:21.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 04:15:09.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 04:29:03.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 04:42:46.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 04:56:27.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:33.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 05:10:10.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:03.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 05:24:03.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.077 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 05:39:29.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:35.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 05:53:10.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:15.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:15.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 06:06:52.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 06:20:40.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:45.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 06:34:21.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:26.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:26.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 06:48:09.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:14.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 07:02:03.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:55.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 07:15:48.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 07:29:32.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:37.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:31.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 07:43:21.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.730 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 07:57:45.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:45.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:52.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 08:11:33.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 08:25:18.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:08.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 08:39:03.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:57.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 08:52:49.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:55.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.542 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 09:06:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 09:20:22.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:23.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:27.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:27.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 09:34:09.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:14.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.159 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 09:47:56.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:01.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:01.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:51.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 10:01:45.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.676 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 10:16:01.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:58.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 10:29:52.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:44.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 10:43:37.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:29.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 10:57:21.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 11:11:08.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:14.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.293 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 11:26:08.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:13.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 11:39:55.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 11:53:37.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:42.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:31.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 12:07:21.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:12.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:05.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 12:21:06.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:07.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:11.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:11.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.914 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 12:35:30.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:37.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:37.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 12:49:16.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:21.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:09.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.064 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 13:03:03.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:08.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:53.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 13:16:44.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:49.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 13:30:31.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.122 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 13:44:15.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 13:58:07.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 14:11:49.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:39.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 14:25:28.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:33.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:25.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 14:39:15.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:21.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:13.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.627 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 14:53:45.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 15:07:27.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 15:21:11.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:16.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 15:34:56.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 15:48:45.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:50.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 16:02:26.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:31.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:31.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:18.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 16:16:06.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:12.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:04.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 16:29:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:30:59.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:30:59.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 16:43:38.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 16:57:20.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:21.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:25.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:25.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:50.805 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 17:12:52.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:47.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 17:26:39.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:44.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 17:40:20.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 17:54:06.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 18:07:49.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 18:21:38.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:37.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 18:35:27.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:32.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:32.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 18:49:15.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:19.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:19.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:05.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 19:03:01.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:07.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 19:16:46.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:51.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:30.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.629 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 19:31:08.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.131 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 19:45:03.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 19:58:46.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:51.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:38.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 20:12:28.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:22.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.523 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 20:26:14.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.131 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 20:40:07.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:11.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:11.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 20:53:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:58.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:54.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 21:07:44.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 21:21:31.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 21:35:16.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:21.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:21.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:11.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.146 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 21:49:43.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 22:03:26.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:32.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 22:17:07.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 22:30:49.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 22:44:32.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 22:59:16.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 23:13:02.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:07.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:07.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:53.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 23:26:44.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 23:40:27.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:26.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-26 23:54:16.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:20.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:20.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.539 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 00:08:40.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 00:22:20.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.490 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 00:36:06.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:59.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 00:49:49.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 01:03:35.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 01:17:20.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:20.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:25.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 01:31:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:11.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:11.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:58.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 01:44:47.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:52.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:52.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 01:58:27.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:14.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 02:12:14.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.197 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 02:26:39.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:45.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:45.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:33.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 02:40:23.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:27.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:27.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 02:54:14.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:09.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 03:08:01.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:06.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:06.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:50.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 03:21:41.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:46.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:47.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 03:35:26.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 03:49:14.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 04:03:03.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:08.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 04:16:48.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:52.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:52.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 04:30:31.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:36.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.602 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 04:45:58.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:04.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:04.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 04:59:46.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:52.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:40.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 05:13:28.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:32.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:32.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.021 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 05:27:07.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:12.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:12.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 05:40:50.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 05:54:35.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:40.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 06:08:19.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:14.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 06:22:07.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:13.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 06:35:50.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:55.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 06:49:30.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.975 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.927 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 07:03:49.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:56.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 07:17:29.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:33.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:33.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:22.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 07:31:16.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:12.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 07:45:07.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 07:58:52.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:57.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 08:12:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:43.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 08:26:20.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:11.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 08:40:03.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:08.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 08:53:47.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:47.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:53.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 09:07:41.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:47.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:47.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.094 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 09:22:03.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 09:35:43.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:48.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 09:49:26.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:31.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:18.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 10:03:08.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:12.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:12.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 10:16:53.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:58.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 10:31:49.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:49.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:55.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 10:45:31.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:32.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 10:59:16.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:21.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:21.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:13.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 11:13:07.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:12.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 11:26:47.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.798 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:41.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.345 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 11:41:06.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:11.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:11.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:59.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.129 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 11:54:48.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:52.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:52.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 12:08:34.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:39.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 12:22:19.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 12:36:01.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 12:49:44.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.549 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 13:03:21.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 13:17:07.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:07.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:12.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:12.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 13:30:47.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:47.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 13:44:38.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:43.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.211 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.375 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 13:58:57.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:02.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 14:12:43.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 14:26:20.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:25.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:12.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:05.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 14:40:05.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:10.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:10.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:58.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 14:53:47.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 15:07:26.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:31.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:31.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.909 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 15:21:06.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:11.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:58.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 15:34:47.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 15:48:32.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:37.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 16:02:14.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:05.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:42.438 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 16:17:43.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:50.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 16:31:26.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 16:45:07.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:59.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 16:58:47.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:52.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:41.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 17:12:31.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 17:26:15.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:20.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:08.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 17:40:00.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:05.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 17:53:45.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:38.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 18:07:25.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:24.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 18:21:14.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:19.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:19.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:07.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.861 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 18:35:38.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:43.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 18:49:21.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 19:03:06.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:11.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 19:16:45.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:50.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:50.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 19:30:30.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:25.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 19:44:14.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:19.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:10.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 19:58:02.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:53.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 20:11:43.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:49.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:39.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 20:25:26.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:19.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 20:39:06.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:12.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.466 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 20:53:29.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:34.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:21.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 21:07:10.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:15.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:06.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 21:21:01.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:51.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 21:34:42.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 21:48:23.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 22:03:21.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:17.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 22:17:06.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:07.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.152 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:57.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 22:30:47.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:52.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 22:44:30.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 22:58:07.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:12.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.698 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 23:12:29.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:35.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:35.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 23:26:15.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:20.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 23:40:02.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-27 23:53:51.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:56.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 00:07:32.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 00:21:21.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:27.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.572 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 00:35:07.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 00:49:02.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:07.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:57.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 01:02:48.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 01:16:33.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.960 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 01:30:58.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 01:44:44.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 01:58:27.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 02:12:08.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:13.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:03.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 02:25:53.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 02:39:38.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:31.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 02:53:20.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:26.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 03:07:08.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:03.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 03:20:53.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:49.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 03:34:40.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:45.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.480 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.493 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 03:50:13.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 04:04:03.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 04:17:48.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:54.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 04:31:40.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 04:45:20.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:11.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 04:59:03.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 05:12:44.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:50.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:35.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 05:26:26.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:22.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 05:40:15.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 05:54:06.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:07.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:12.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:25.178 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 06:08:27.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:21.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 06:22:11.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:17.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:07.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 06:36:02.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:07.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 06:49:45.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 07:03:27.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 07:17:12.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:17.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:04.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 07:30:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:48.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 07:44:40.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:33.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 07:58:21.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:26.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:26.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 08:12:08.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:13.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.518 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 08:26:31.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 08:40:22.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:27.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:27.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:13.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 08:54:08.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:13.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 08:57:00.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 09:07:50.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:51.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:56.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 09:21:32.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:38.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:38.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 09:36:35.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 09:50:22.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:24.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.615 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 10:04:16.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 10:18:09.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:14.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:02.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 10:31:54.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:53.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.609 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 10:46:20.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 11:00:08.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 11:13:51.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 11:27:32.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 11:41:15.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 11:55:03.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 12:08:48.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:53.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:47.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 12:22:40.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 12:36:24.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 12:50:08.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:09.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:13.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:13.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.769 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 13:04:33.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:33.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 13:18:22.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:27.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:27.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:14.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 13:32:07.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 13:45:50.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:55.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.544 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 13:59:30.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:35.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:35.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:21.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 14:13:08.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 14:27:02.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 14:40:44.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:48.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:48.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 14:54:25.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 15:08:07.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:12.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:02.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.434 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 15:23:36.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:32.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.373 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 15:37:21.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 15:51:08.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 16:04:54.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:50.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 16:18:40.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:46.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 16:32:23.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:28.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 16:46:07.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 16:59:50.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 17:13:34.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 17:27:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:26.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.601 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 17:41:44.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:49.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:49.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 17:55:31.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 18:09:15.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:21.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 18:23:03.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 18:36:44.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:49.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:49.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:39.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 18:50:26.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:22.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 19:04:13.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:14.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:19.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.109 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:08.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.129 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 19:18:02.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:06.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:06.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 19:31:44.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.662 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 19:45:25.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.549 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.363 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 19:59:44.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 20:13:24.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:30.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:15.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 20:27:07.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:12.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 20:40:49.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:54.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 20:54:36.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 21:09:41.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:47.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 21:23:23.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:29.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.845 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 21:37:07.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 21:50:48.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:54.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 22:04:31.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:31.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:37.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.757 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 22:18:53.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 22:32:40.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:35.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 22:46:24.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:29.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:06.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 23:00:07.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:59.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 23:13:49.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:50.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:54.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 23:27:32.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:23.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 23:41:14.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-28 23:55:00.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-29 00:08:39.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:29.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-29 00:22:19.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.260 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-29 00:36:43.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:47.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:47.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-29 00:50:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 19, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 19, dp_rank 9: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=4, rank_in_group=1 +2026-06-29 01:04:09.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 19 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:14.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 19 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 19: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:04.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 19: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 19: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 19, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_22.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_22.log new file mode 100644 index 0000000000000000000000000000000000000000..ad3dfe6160d85c5baa8f2d688c24e77730cbb430 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_22.log @@ -0,0 +1,6079 @@ +2026-05-29 20:46:15.583 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-29 20:46:15.585 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-29 20:46:16.276 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:46:16.277 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 20:51:46.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_5/index_2550_seed_24566_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 20:53:30.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:30.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:38.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_5/index_2550_seed_24567_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 20:55:17.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 22, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 21:06:57.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:04.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_5/index_5756_seed_18818_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 21:08:17.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:17.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:09:39.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_5/index_5756_seed_18819_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:10:01.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.122 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 21:21:17.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_5/index_267_seed_25097_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 21:22:43.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:43.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:23:53.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_5/index_267_seed_25098_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:24:15.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 21:35:35.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_5/index_4265_seed_16042_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 21:37:04.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:37:04.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:13.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_5/index_4265_seed_16043_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:38:36.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 21:49:56.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_5/index_3689_seed_15733_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 21:51:22.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:22.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:52:35.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_5/index_3689_seed_15734_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:53:01.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.131 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 22:04:18.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_5/index_8820_seed_33040_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 22:05:40.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:40.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:06:52.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_5/index_8820_seed_33041_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:07:15.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 22:18:35.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_5/index_6003_seed_32401_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 22:19:56.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:56.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:14.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_5/index_6003_seed_32402_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:21:38.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 22:32:58.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:58.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_5/index_53_seed_13991_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 22:34:21.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:21.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:35.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_5/index_53_seed_13992_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:36:01.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.542 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.189 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 22:47:20.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_5/index_7232_seed_8916_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 22:48:44.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:44.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:49:56.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_5/index_7232_seed_8917_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:50:20.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 23:01:44.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_5/index_1416_seed_28603_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 23:03:08.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:03:08.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:23.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_5/index_1416_seed_28604_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:04:49.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:52.861 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 23:16:57.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_5/index_8218_seed_33473_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 23:18:19.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:19.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:35.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_5/index_8218_seed_33474_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:19:57.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 23:31:14.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_5/index_4083_seed_16330_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 23:32:36.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:36.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:33:52.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_5/index_4083_seed_16331_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:34:15.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.142 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 23:45:29.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_5/index_5673_seed_30570_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-29 23:46:51.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:51.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:08.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_5/index_5673_seed_30571_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:48:31.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-29 23:59:50.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_5/index_94_seed_12267_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 00:01:20.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:01:20.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:26.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_5/index_94_seed_12268_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:02:55.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 00:14:13.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:20.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_5/index_3576_seed_2077_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 00:15:37.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:37.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:16:51.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_5/index_3576_seed_2078_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:17:18.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 00:28:36.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_5/index_7715_seed_3587_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 00:29:58.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:58.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:16.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_5/index_7715_seed_3588_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:31:39.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 00:42:58.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_5/index_4347_seed_28621_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 00:44:24.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:24.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:45:37.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_5/index_4347_seed_28622_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:46:02.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.142 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 00:57:15.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_5/index_559_seed_19790_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 00:58:38.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:38.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:59:51.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_5/index_559_seed_19791_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:00:16.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 01:11:38.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:45.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_5/index_8683_seed_5589_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 01:13:01.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:13:01.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:18.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_5/index_8683_seed_5590_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:14:44.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 01:26:05.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_5/index_4475_seed_13155_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 01:27:28.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:28.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:28:41.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_5/index_4475_seed_13156_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 01:29:03.428 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:29:04.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.608 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 01:41:19.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_5/index_8541_seed_11454_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 01:42:43.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:43.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:43:56.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_5/index_8541_seed_11455_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:44:18.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 01:55:38.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_5/index_3252_seed_31890_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 01:56:58.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:58.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:14.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_5/index_3252_seed_31891_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:58:39.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 02:09:58.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:05.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_5/index_9209_seed_26458_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 02:11:18.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:18.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:12:39.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_5/index_9209_seed_26459_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:13:02.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 02:24:23.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_5/index_3170_seed_16942_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 02:25:49.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:49.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:05.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_5/index_3170_seed_16943_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:27:33.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 02:38:56.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_5/index_1683_seed_33164_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 02:40:18.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:18.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:34.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_5/index_1683_seed_33165_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:41:58.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 02:54:31.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_5/index_5682_seed_20368_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 02:55:52.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:52.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:08.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_5/index_5682_seed_20369_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:57:31.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 03:08:50.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_5/index_2446_seed_28432_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 03:10:11.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:10:11.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:26.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_5/index_2446_seed_28433_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:11:53.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 03:23:09.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:17.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_5/index_4364_seed_12852_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 03:24:31.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:31.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:25:46.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_5/index_4364_seed_12853_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:26:08.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 03:37:26.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_5/index_5398_seed_11883_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 03:38:48.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:48.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:07.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_5/index_5398_seed_11884_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:40:39.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 03:51:59.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_5/index_7104_seed_21270_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 03:53:25.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:25.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:54:38.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_5/index_7104_seed_21271_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:55:04.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.584 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 04:07:19.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:28.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_5/index_5925_seed_28754_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 04:08:43.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:43.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:01.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_5/index_5925_seed_28755_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:10:28.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 04:21:48.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_5/index_7437_seed_33473_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 04:23:14.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:23:14.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:28.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_5/index_7437_seed_33474_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:24:54.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 04:36:12.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_5/index_3670_seed_6601_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 04:37:33.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:33.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:38:50.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_5/index_3670_seed_6602_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:39:16.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 04:50:41.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:48.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_5/index_457_seed_4304_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 04:52:03.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:52:03.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:19.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_5/index_457_seed_4305_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:53:44.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 05:05:07.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_5/index_2799_seed_22047_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 05:06:29.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:29.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:07:43.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_5/index_2799_seed_22048_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:08:13.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 05:19:25.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_5/index_8378_seed_4538_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 05:20:46.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:46.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:21:58.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_5/index_8378_seed_4539_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:22:20.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 05:33:38.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:45.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_5/index_1077_seed_28825_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 05:35:05.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:35:05.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:15.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_5/index_1077_seed_28826_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:36:37.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 05:47:59.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:07.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_5/index_726_seed_4758_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 05:49:24.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:24.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:35.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_5/index_726_seed_4759_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:50:58.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 06:02:21.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:30.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_5/index_8468_seed_9429_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 06:03:46.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:46.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:04.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_5/index_8468_seed_9430_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:05:33.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 06:16:55.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:03.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_5/index_4402_seed_18699_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 06:18:17.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:17.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:31.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_5/index_4402_seed_18700_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:19:59.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:11.809 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 06:32:17.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_5/index_9606_seed_10585_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 06:33:37.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:37.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:34:55.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_5/index_9606_seed_10586_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:35:16.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 06:46:36.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_5/index_8519_seed_18860_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 06:47:59.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:59.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:14.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_5/index_8519_seed_18861_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:49:37.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 07:00:59.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_5/index_8745_seed_21220_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 07:02:21.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:21.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:37.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_5/index_8745_seed_21221_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:04:00.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 07:15:21.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:29.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_5/index_7086_seed_16039_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 07:16:43.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:43.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:06.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_5/index_7086_seed_16040_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:18:41.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 07:30:08.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_5/index_5082_seed_23722_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 07:31:36.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:36.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:32:48.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_5/index_5082_seed_23723_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:33:11.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.021 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 07:44:30.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_5/index_7320_seed_18632_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 07:45:58.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:58.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:04.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_5/index_7320_seed_18633_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:47:30.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 07:59:00.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_5/index_243_seed_16963_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 08:00:22.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:22.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:01:41.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_5/index_243_seed_16964_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:02:08.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 08:13:34.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_5/index_5948_seed_26952_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 08:14:55.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:55.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:17.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_5/index_5948_seed_26953_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:16:44.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 08:28:15.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_5/index_9652_seed_25844_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 08:29:43.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:43.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:30:55.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_5/index_9652_seed_25845_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 08:31:22.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:31:23.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 08:42:53.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_5/index_9887_seed_3952_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 08:44:16.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:16.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:45:35.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_5/index_9887_seed_3953_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:46:01.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:24.636 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 08:59:29.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_5/index_8182_seed_25086_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 09:00:54.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:54.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:11.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_5/index_8182_seed_25087_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:02:32.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 09:14:05.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:12.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_5/index_9386_seed_16459_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 09:15:26.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:26.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:16:52.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_5/index_9386_seed_16460_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:17:17.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 09:28:42.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:43.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_5/index_8292_seed_2435_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 09:30:04.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:30:04.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:27.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_5/index_8292_seed_2436_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:31:54.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 09:43:17.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:26.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_5/index_5561_seed_32480_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 09:44:40.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:40.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:02.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_5/index_5561_seed_32481_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 09:46:24.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:46:25.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 09:57:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:54.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_5/index_5659_seed_21341_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 09:59:08.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:59:08.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:23.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_5/index_5659_seed_21342_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:00:52.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 10:12:07.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:16.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_5/index_9189_seed_11100_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 10:13:31.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:31.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:14:44.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_5/index_9189_seed_11101_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:15:06.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 10:26:24.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_5/index_1652_seed_8375_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 10:27:47.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:47.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:02.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_5/index_1652_seed_8376_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:29:29.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 10:40:46.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_5/index_435_seed_18082_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 10:42:07.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:42:07.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:28.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_5/index_435_seed_18083_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:43:55.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 10:55:08.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_5/index_9757_seed_24743_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 10:56:31.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:31.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:57:52.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_5/index_9757_seed_24744_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:58:23.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 11:09:40.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_5/index_6531_seed_9182_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 11:11:04.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:11:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:20.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_5/index_6531_seed_9183_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:12:43.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:03.095 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 11:25:08.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_5/index_3553_seed_24618_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 11:26:33.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:33.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:27:52.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_5/index_3553_seed_24619_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:28:21.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 11:39:42.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_5/index_887_seed_19170_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 11:41:03.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:41:03.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:19.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_5/index_887_seed_19171_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:42:42.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 11:54:06.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_5/index_6095_seed_20669_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 11:55:30.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:30.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:56:41.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_5/index_6095_seed_20670_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:57:06.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 12:08:30.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_5/index_8535_seed_14671_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 12:09:53.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:53.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:07.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_5/index_8535_seed_14672_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:11:30.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 12:22:53.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_5/index_3357_seed_32580_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 12:24:20.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:20.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:30.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_5/index_3357_seed_32581_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:25:55.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 12:37:12.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_5/index_9855_seed_16684_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 12:38:34.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:34.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:39:47.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_5/index_9855_seed_16685_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:40:10.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 12:51:30.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_5/index_1949_seed_11998_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 12:52:51.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:51.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:11.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_5/index_1949_seed_11999_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:54:32.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 13:05:59.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:07.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_5/index_8576_seed_29600_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 13:07:22.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:22.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:08:38.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_5/index_8576_seed_29601_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:09:04.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.293 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 13:20:24.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:32.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_5/index_5640_seed_9438_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 13:21:48.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:48.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:00.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_5/index_5640_seed_9439_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:23:30.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 13:34:48.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_5/index_7565_seed_33215_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 13:36:11.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:36:11.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:25.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_5/index_7565_seed_33216_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:37:48.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.608 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 13:50:02.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_5/index_6362_seed_31590_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 13:51:32.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:32.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:52:48.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_5/index_6362_seed_31591_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:53:11.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 14:04:33.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:34.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:44.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_5/index_7328_seed_4776_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 14:05:59.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:59.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:19.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_5/index_7328_seed_4777_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:07:59.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 14:19:43.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_5/index_6309_seed_6100_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 14:21:09.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:21:09.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:22:38.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_5/index_6309_seed_6101_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:23:13.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 14:34:38.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:46.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_5/index_2128_seed_15071_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 14:35:59.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:59.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:16.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_5/index_2128_seed_15072_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:37:54.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 14:49:20.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:28.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_5/index_6634_seed_33416_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 14:50:40.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:40.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:00.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_5/index_6634_seed_33417_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:52:26.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 15:05:01.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_5/index_316_seed_24288_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 15:06:24.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:24.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:07:40.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_5/index_316_seed_24289_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:08:04.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 15:19:35.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_5/index_5641_seed_28684_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 15:20:58.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:58.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:14.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_5/index_5641_seed_28685_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:22:38.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 15:34:11.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_5/index_9255_seed_32217_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 15:35:40.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:40.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:36:53.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_5/index_9255_seed_32218_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:37:19.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 15:48:48.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_5/index_1767_seed_16655_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 15:50:16.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:50:16.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_5/index_1767_seed_16656_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:51:51.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 16:03:17.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_5/index_7923_seed_27657_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 16:04:43.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:43.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:05:57.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_5/index_7923_seed_27658_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:06:20.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.687 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 16:18:48.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_5/index_2140_seed_3275_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 16:20:17.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:17.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:21:33.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_5/index_2140_seed_3276_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:22:05.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 16:33:33.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:33.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:43.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_5/index_170_seed_23162_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 16:35:00.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:35:00.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:16.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_5/index_170_seed_23163_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:36:40.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 16:48:15.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:24.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_5/index_720_seed_21083_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 16:49:38.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:38.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:50:53.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_5/index_720_seed_21084_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:51:16.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 17:02:55.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_5/index_2572_seed_13747_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 17:04:18.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:18.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:33.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_5/index_2572_seed_13748_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:05:57.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 17:17:35.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_5/index_8989_seed_715_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 17:18:58.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:58.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:15.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_5/index_8989_seed_716_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:20:38.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 17:32:10.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_5/index_7207_seed_19069_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 17:33:37.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:37.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:00.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_5/index_7207_seed_19070_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:35:26.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 17:46:54.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_5/index_5553_seed_20816_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 17:48:18.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:18.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:49:38.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_5/index_5553_seed_20817_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:50:02.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 18:01:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_5/index_5234_seed_15256_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 18:03:04.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:03:04.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:17.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_5/index_5234_seed_15257_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:04:40.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 18:16:18.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_5/index_7830_seed_31949_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 18:17:43.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:43.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:00.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_5/index_7830_seed_31950_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:19:32.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 18:31:18.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:30.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_5/index_5562_seed_21165_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 18:32:43.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:43.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:14.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_5/index_5562_seed_21166_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:34:46.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:17.861 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 18:47:24.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_5/index_4017_seed_3701_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 18:48:48.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:48.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:12.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_5/index_4017_seed_3702_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:50:39.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 19:02:13.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_5/index_5199_seed_7497_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 19:03:39.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:39.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:00.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_5/index_5199_seed_7498_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:05:34.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 19:17:12.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:22.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_5/index_1810_seed_2412_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 19:18:37.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:37.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:19:58.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_5/index_1810_seed_2413_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:20:26.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 19:31:55.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_5/index_3811_seed_28315_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 19:33:23.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:23.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:34:55.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_5/index_3811_seed_28316_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:35:24.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 19:47:03.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_5/index_7688_seed_5847_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 19:48:25.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:25.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:49:47.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_5/index_7688_seed_5848_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:50:11.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 20:01:46.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_5/index_4168_seed_26691_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 20:03:10.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:03:10.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:29.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_5/index_4168_seed_26692_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:04:56.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 20:16:29.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_5/index_1245_seed_5037_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 20:17:55.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:55.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:19.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_5/index_1245_seed_5038_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:19:47.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 20:31:20.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:29.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_5/index_7658_seed_11944_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 20:32:43.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:43.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:00.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_5/index_7658_seed_11945_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:34:24.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 20:46:03.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_5/index_4831_seed_23750_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 20:47:25.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:25.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:48:49.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_5/index_4831_seed_23751_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:49:15.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 21:00:47.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:47.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_5/index_3483_seed_33086_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 21:02:16.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:02:16.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:31.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_5/index_3483_seed_33087_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:03:58.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.370 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 21:17:31.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:41.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_5/index_9503_seed_26261_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 21:18:59.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:59.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:12.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_5/index_9503_seed_26262_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:20:37.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 21:32:09.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:49:13.996 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-30 21:49:13.996 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-30 21:49:14.672 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:49:14.673 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 21:55:22.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:49.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_5/index_2550_seed_24566_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 21:57:09.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:57:09.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:18.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_5/index_2550_seed_24567_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:59:07.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 22, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 22:10:52.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:12:00.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_5/index_5756_seed_18818_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 22:12:18.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:18.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:13:35.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_5/index_5756_seed_18819_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:14:03.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 22:25:33.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:42.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_5/index_267_seed_25097_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 22:26:55.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:55.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:14.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_5/index_267_seed_25098_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:28:37.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 22:40:03.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_5/index_4265_seed_16042_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 22:41:32.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:32.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:42:45.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_5/index_4265_seed_16043_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:43:10.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 22:54:38.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_5/index_3689_seed_15733_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 22:56:03.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:56:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:25.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_5/index_3689_seed_15734_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:57:50.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 23:09:20.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_5/index_8820_seed_33040_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 23:10:41.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:41.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:11.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_5/index_8820_seed_33041_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:12:36.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 23:24:05.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:14.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_5/index_6003_seed_32401_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 23:25:31.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:31.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:26:47.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_5/index_6003_seed_32402_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:27:14.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 23:38:46.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:55.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_5/index_53_seed_13991_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 23:40:10.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:40:10.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:25.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_5/index_53_seed_13992_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:41:48.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-30 23:53:10.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_5/index_7232_seed_8916_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-30 23:54:32.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:32.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:55:52.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_5/index_7232_seed_8917_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:56:15.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 00:07:41.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_5/index_1416_seed_28603_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 00:09:11.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:09:11.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:24.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_5/index_1416_seed_28604_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:10:46.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:54.355 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 00:22:59.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_5/index_8218_seed_33473_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 00:24:24.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:24.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:25:40.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_5/index_8218_seed_33474_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:26:05.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 00:37:30.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:40.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_5/index_4083_seed_16330_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 00:38:56.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:56.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:18.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_5/index_4083_seed_16331_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:40:46.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 00:52:07.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_5/index_5673_seed_30570_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 00:53:34.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:34.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:54:48.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_5/index_5673_seed_30571_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:55:22.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 01:06:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_5/index_94_seed_12267_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 01:08:11.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:08:11.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:26.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_5/index_94_seed_12268_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:09:52.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 01:21:16.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_5/index_3576_seed_2077_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 01:22:42.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:42.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:23:56.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_5/index_3576_seed_2078_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:24:20.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 01:35:36.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_5/index_7715_seed_3587_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 01:36:57.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:57.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:15.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_5/index_7715_seed_3588_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:38:38.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 01:50:03.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_5/index_4347_seed_28621_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 01:51:29.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:29.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:52:42.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_5/index_4347_seed_28622_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:53:06.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 02:04:34.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_5/index_559_seed_19790_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 02:06:01.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:06:01.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:13.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_5/index_559_seed_19791_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:07:39.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 22, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-05-31 02:19:01.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_5/index_8683_seed_5589_dp_11_rank_22_sample_2_slot_0.mp4 +2026-05-31 02:20:21.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:21.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:21:38.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_5/index_8683_seed_5590_dp_11_rank_22_sample_3_slot_1.mp4 +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:22:02.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:25.030 | INFO | __main__:__init__:946 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-06-25 18:01:32.162 | INFO | __main__:__init__:947 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-06-25 18:01:32.852 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:32.879 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 18:04:09.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_5/index_2550_seed_24566_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 18:05:51.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:51.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:06:56.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_5/index_2550_seed_24567_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:07:32.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 22, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 18:18:41.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:47.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_5/index_5756_seed_18818_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 18:20:03.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:20:03.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:12.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_5/index_5756_seed_18819_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:21:35.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 18:32:26.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:32.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_5/index_267_seed_25097_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 18:33:49.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:49.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:34:54.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_5/index_267_seed_25098_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:35:15.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 18:46:08.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_5/index_4265_seed_16042_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 18:47:32.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:32.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:48:38.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_5/index_4265_seed_16043_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:49:03.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 19:00:02.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:02.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_5/index_3689_seed_15733_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 19:01:25.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:25.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:34.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_5/index_3689_seed_15734_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:02:57.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 19:13:49.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_5/index_8820_seed_33040_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 19:15:09.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:15:09.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:20.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_5/index_8820_seed_33041_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:16:42.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 19:27:32.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_5/index_6003_seed_32401_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 19:28:54.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:54.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:03.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_5/index_6003_seed_32402_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:30:24.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 19:41:16.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_5/index_53_seed_13991_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 19:42:39.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:39.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:43:50.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_5/index_53_seed_13992_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:44:15.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 19:55:08.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:13.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_5/index_7232_seed_8916_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 19:56:28.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:28.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:57:41.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_5/index_7232_seed_8917_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:58:03.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 20:09:02.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:08.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_5/index_1416_seed_28603_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 20:10:27.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:27.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:35.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_5/index_1416_seed_28604_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:12:00.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.666 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 20:23:32.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_5/index_8218_seed_33473_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 20:24:52.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:52.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:05.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_5/index_8218_seed_33474_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:26:26.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 20:37:18.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:25.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_5/index_4083_seed_16330_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 20:38:39.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:39.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:39:48.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_5/index_4083_seed_16331_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:40:09.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 20:51:03.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:09.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_5/index_5673_seed_30570_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 20:52:28.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:28.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:35.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_5/index_5673_seed_30571_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:53:57.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 21:04:50.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:51.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:57.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_5/index_94_seed_12267_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 21:06:18.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:06:18.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:23.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_5/index_94_seed_12268_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:07:49.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 21:18:43.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:49.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_5/index_3576_seed_2077_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 21:20:07.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:20:07.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:17.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_5/index_3576_seed_2078_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:21:39.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 21:32:31.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_5/index_7715_seed_3587_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 21:33:50.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:50.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:01.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_5/index_7715_seed_3588_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:35:21.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 21:46:16.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:22.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_5/index_4347_seed_28621_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 21:47:37.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:37.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:48:47.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_5/index_4347_seed_28622_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:49:10.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 22:00:05.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_5/index_559_seed_19790_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 22:01:27.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:27.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:36.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_5/index_559_seed_19791_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:02:58.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 22:13:50.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_5/index_8683_seed_5589_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 22:15:11.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:15:11.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:21.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_5/index_8683_seed_5590_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:16:45.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 22:27:39.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_5/index_4475_seed_13155_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 22:29:00.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:29:00.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:10.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_5/index_4475_seed_13156_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:30:32.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:57.631 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 22:41:59.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:41:59.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:07.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_5/index_8541_seed_11454_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 22:43:21.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:21.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:31.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_5/index_8541_seed_11455_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:44:52.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 22:55:45.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:51.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_5/index_3252_seed_31890_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 22:57:05.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:57:05.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:16.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_5/index_3252_seed_31891_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:58:39.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 23:09:30.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_5/index_9209_seed_26458_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 23:10:50.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:50.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:01.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_5/index_9209_seed_26459_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:12:22.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 23:23:15.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_5/index_3170_seed_16942_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 23:24:37.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:37.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:25:47.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_5/index_3170_seed_16943_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:26:11.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 23:37:07.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_5/index_1683_seed_33164_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 23:38:27.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:27.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:39:37.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_5/index_1683_seed_33165_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:40:02.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-25 23:51:52.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_5/index_5682_seed_20368_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-25 23:53:12.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:53:12.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:22.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_5/index_5682_seed_20369_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:54:43.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 00:05:36.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_5/index_2446_seed_28432_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 00:06:57.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:57.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:07.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_5/index_2446_seed_28433_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:08:28.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 00:19:21.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_5/index_4364_seed_12852_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 00:20:41.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:41.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:21:50.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_5/index_4364_seed_12853_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:22:11.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 00:33:07.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:13.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_5/index_5398_seed_11883_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 00:34:27.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:27.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:35:37.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_5/index_5398_seed_11884_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:36:04.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 00:47:03.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:09.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_5/index_7104_seed_21270_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 00:48:23.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:23.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:38.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_5/index_7104_seed_21271_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:50:00.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:29.056 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 01:01:30.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:31.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:38.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_5/index_5925_seed_28754_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 01:02:51.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:51.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:04.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_5/index_5925_seed_28755_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:04:25.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 01:15:17.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:24.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_5/index_7437_seed_33473_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 01:16:40.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:40.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:17:50.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_5/index_7437_seed_33474_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:18:11.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 01:29:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:14.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_5/index_3670_seed_6601_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 01:30:29.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:29.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:31:40.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_5/index_3670_seed_6602_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:32:03.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 01:42:58.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:04.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_5/index_457_seed_4304_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 01:44:20.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:20.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:29.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_5/index_457_seed_4305_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:45:50.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 01:56:44.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:50.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_5/index_2799_seed_22047_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 01:58:06.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:58:06.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:17.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_5/index_2799_seed_22048_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:59:40.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 02:10:31.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:37.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_5/index_8378_seed_4538_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 02:11:52.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:52.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:03.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_5/index_8378_seed_4539_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:13:25.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 02:24:15.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_5/index_1077_seed_28825_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 02:25:36.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:36.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:26:46.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_5/index_1077_seed_28826_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:27:11.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 02:38:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:14.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_5/index_726_seed_4758_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 02:39:29.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:29.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:37.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_5/index_726_seed_4759_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 02:40:59.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:41:00.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 02:51:53.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:54.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_5/index_8468_seed_9429_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 02:53:16.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:53:16.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:25.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_5/index_8468_seed_9430_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:54:48.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 03:05:40.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:41.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_5/index_4402_seed_18699_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 03:07:00.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:07:00.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:14.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_5/index_4402_seed_18700_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:08:38.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:05.061 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 03:20:06.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_5/index_9606_seed_10585_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 03:21:25.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:25.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:39.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_5/index_9606_seed_10586_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:22:59.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 03:33:52.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_5/index_8519_seed_18860_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 03:35:14.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:35:14.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:23.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_5/index_8519_seed_18861_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:36:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 03:47:40.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_5/index_8745_seed_21220_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 03:48:58.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:58.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:10.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_5/index_8745_seed_21221_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:50:31.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 04:01:21.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:27.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_5/index_7086_seed_16039_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 04:02:40.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:40.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:03:51.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_5/index_7086_seed_16040_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:04:12.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 04:15:09.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_5/index_5082_seed_23722_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 04:16:33.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:33.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:17:42.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_5/index_5082_seed_23723_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:18:05.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 04:29:03.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_5/index_7320_seed_18632_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 04:30:24.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:24.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:33.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_5/index_7320_seed_18633_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:31:53.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 04:42:46.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:52.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_5/index_243_seed_16963_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 04:44:07.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:44:07.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:16.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_5/index_243_seed_16964_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:45:35.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 04:56:27.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:32.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_5/index_5948_seed_26952_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 04:57:46.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:46.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:58:58.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_5/index_5948_seed_26953_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:59:19.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 05:10:10.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_5/index_9652_seed_25844_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 05:11:35.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:35.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:12:41.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_5/index_9652_seed_25845_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:13:03.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 05:24:03.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:09.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_5/index_9887_seed_3952_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 05:25:23.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:23.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:34.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_5/index_9887_seed_3953_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:26:58.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.340 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 05:39:29.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_5/index_8182_seed_25086_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 05:40:50.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:50.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:00.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_5/index_8182_seed_25087_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:42:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 05:53:10.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:16.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_5/index_9386_seed_16459_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 05:54:32.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:32.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:39.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_5/index_9386_seed_16460_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:56:00.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 06:06:52.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:58.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_5/index_8292_seed_2435_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 06:08:12.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:08:12.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:22.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_5/index_8292_seed_2436_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:09:44.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 06:20:40.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_5/index_5561_seed_32480_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 06:22:01.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:22:01.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:09.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_5/index_5561_seed_32481_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:23:31.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 06:34:21.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:22.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_5/index_5659_seed_21341_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 06:35:44.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:44.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:36:51.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_5/index_5659_seed_21342_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:37:11.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 06:48:09.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_5/index_9189_seed_11100_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 06:49:30.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:30.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:50:44.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_5/index_9189_seed_11101_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:51:03.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.281 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 07:02:03.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_5/index_1652_seed_8375_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 07:03:26.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:26.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:33.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_5/index_1652_seed_8376_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:04:55.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 07:15:48.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:54.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_5/index_435_seed_18082_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 07:17:08.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:17:08.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:21.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_5/index_435_seed_18083_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:18:41.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 07:29:32.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:38.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_5/index_9757_seed_24743_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 07:30:53.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:53.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:04.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_5/index_9757_seed_24744_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:32:30.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 07:43:21.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:27.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_5/index_6531_seed_9182_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 07:44:48.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:48.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:45:53.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_5/index_6531_seed_9183_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:46:15.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:44.036 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 07:57:45.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:45.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:53.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_5/index_3553_seed_24618_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 07:59:08.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:59:08.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:18.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_5/index_3553_seed_24619_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:00:40.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 08:11:33.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_5/index_887_seed_19170_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 08:12:55.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:55.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:02.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_5/index_887_seed_19171_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:14:26.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 08:25:18.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:24.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_5/index_6095_seed_20669_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 08:26:41.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:41.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:27:47.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_5/index_6095_seed_20670_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:28:08.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 08:39:03.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:09.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_5/index_8535_seed_14671_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 08:40:26.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:26.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:35.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_5/index_8535_seed_14672_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:41:57.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 08:52:49.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_5/index_3357_seed_32580_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 08:54:11.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:54:11.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:21.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_5/index_3357_seed_32581_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:55:45.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 09:06:39.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:45.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_5/index_9855_seed_16684_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 09:07:59.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:59.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:11.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_5/index_9855_seed_16685_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:09:33.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 09:20:22.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:28.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_5/index_1949_seed_11998_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 09:21:43.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:43.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:22:52.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_5/index_1949_seed_11999_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:23:12.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 09:34:09.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_5/index_8576_seed_29600_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 09:35:32.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:32.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:36:41.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_5/index_8576_seed_29601_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:37:04.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 09:47:56.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:57.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_5/index_5640_seed_9438_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 09:49:18.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:18.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:29.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_5/index_5640_seed_9439_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:50:51.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 10:01:45.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_5/index_7565_seed_33215_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 10:03:05.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:03:05.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:14.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_5/index_7565_seed_33216_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:04:34.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.713 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 10:16:01.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:02.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:09.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_5/index_6362_seed_31590_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 10:17:29.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:29.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:34.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_5/index_6362_seed_31591_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:18:59.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 10:29:52.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:58.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_5/index_7328_seed_4776_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 10:31:12.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:31:12.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:23.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_5/index_7328_seed_4777_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:32:44.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 10:43:37.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:43.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_5/index_6309_seed_6100_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 10:44:57.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:57.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_5/index_6309_seed_6101_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:46:29.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 10:57:21.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_5/index_2128_seed_15071_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 10:58:43.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:43.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:59:56.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_5/index_2128_seed_15072_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:00:18.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 11:11:08.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:09.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_5/index_6634_seed_33416_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 11:12:29.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:29.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:38.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_5/index_6634_seed_33417_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:13:59.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.289 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 11:26:08.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_5/index_316_seed_24288_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 11:27:29.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:29.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:28:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_5/index_316_seed_24289_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 11:29:02.143 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:29:03.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 11:39:55.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_5/index_5641_seed_28684_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 11:41:15.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:15.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:24.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_5/index_5641_seed_28685_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:42:44.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 11:53:37.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_5/index_9255_seed_32217_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 11:55:02.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:55:02.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:07.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_5/index_9255_seed_32218_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:56:31.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 12:07:21.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_5/index_1767_seed_16655_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 12:08:46.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:46.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:09:51.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_5/index_1767_seed_16656_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:10:12.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 12:21:06.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_5/index_7923_seed_27657_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 12:22:28.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:28.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:38.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_5/index_7923_seed_27658_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 12:23:59.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:24:00.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.986 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 12:35:30.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:38.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_5/index_2140_seed_3275_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 12:36:53.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:53.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:03.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_5/index_2140_seed_3276_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:38:25.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 12:49:16.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_5/index_170_seed_23162_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 12:50:40.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:40.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:51:47.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_5/index_170_seed_23163_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:52:10.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 13:03:03.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_5/index_720_seed_21083_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 13:04:22.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:22.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:33.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_5/index_720_seed_21084_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:05:53.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 13:16:44.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_5/index_2572_seed_13747_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 13:18:03.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:18:03.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:16.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_5/index_2572_seed_13748_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:19:41.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 13:30:31.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_5/index_8989_seed_715_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 13:31:51.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:51.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:04.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_5/index_8989_seed_716_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:33:25.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 13:44:15.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:21.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_5/index_7207_seed_19069_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 13:45:40.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:40.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:46:49.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_5/index_7207_seed_19070_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:47:13.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 13:58:07.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:13.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_5/index_5553_seed_20816_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 13:59:27.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:27.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:36.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_5/index_5553_seed_20817_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:00:56.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 14:11:49.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:55.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_5/index_5234_seed_15256_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 14:13:09.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:13:09.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:17.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_5/index_5234_seed_15257_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:14:39.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 14:25:28.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_5/index_7830_seed_31949_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 14:26:57.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:57.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:02.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_5/index_7830_seed_31950_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:28:25.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 14:39:15.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_5/index_5562_seed_21165_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 14:40:35.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:35.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:41:50.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_5/index_5562_seed_21166_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:42:13.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.693 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 14:53:45.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_5/index_4017_seed_3701_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 14:55:06.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:55:06.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:16.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_5/index_4017_seed_3702_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:56:36.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 15:07:27.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_5/index_5199_seed_7497_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 15:08:47.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:47.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:09:56.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_5/index_5199_seed_7498_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:10:19.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 15:21:10.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_5/index_1810_seed_2412_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 15:22:32.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:32.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:23:41.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_5/index_1810_seed_2413_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:24:02.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 15:34:56.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_5/index_3811_seed_28315_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 15:36:19.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:19.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:29.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_5/index_3811_seed_28316_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:37:50.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 15:48:45.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_5/index_7688_seed_5847_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 15:50:03.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:50:03.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:15.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_5/index_7688_seed_5848_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:51:35.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 16:02:26.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_5/index_4168_seed_26691_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 16:03:45.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:45.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:04:57.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_5/index_4168_seed_26692_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:05:18.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 16:16:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:07.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_5/index_1245_seed_5037_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 16:17:27.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:27.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:18:42.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_5/index_1245_seed_5038_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:19:04.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 16:29:54.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_5/index_7658_seed_11944_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 16:31:14.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:14.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:23.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_5/index_7658_seed_11945_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 16:32:44.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:32:45.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.122 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 16:43:38.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_5/index_4831_seed_23750_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 16:44:57.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:57.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_5/index_4831_seed_23751_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:46:31.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 16:57:20.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_5/index_3483_seed_33086_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 16:58:48.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:48.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:59:54.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_5/index_3483_seed_33087_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:00:16.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.228 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 17:12:53.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_5/index_9503_seed_26261_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 17:14:18.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:14:18.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:25.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_5/index_9503_seed_26262_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:15:48.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 17:26:39.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:45.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_5/index_599_seed_169_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 17:28:00.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:28:00.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:09.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_5/index_599_seed_170_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:29:30.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 17:40:20.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:26.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_5/index_6975_seed_30331_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 17:41:40.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:40.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:42:54.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_5/index_6975_seed_30332_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:43:14.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 17:54:06.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_5/index_1874_seed_3053_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 17:55:28.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:28.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:37.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_5/index_1874_seed_3054_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:56:58.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 18:07:49.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:55.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_5/index_8392_seed_26475_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 18:09:12.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:09:12.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:21.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_5/index_8392_seed_26476_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:10:45.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 18:21:38.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_5/index_7847_seed_20032_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 18:22:58.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:58.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_5/index_7847_seed_20033_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:24:37.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 18:35:27.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_5/index_4741_seed_9_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 18:36:53.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:53.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:03.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_5/index_4741_seed_10_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:38:24.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 18:49:15.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_5/index_2838_seed_9611_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 18:50:36.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:36.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:51:45.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_5/index_2838_seed_9612_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:52:05.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 19:03:01.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:01.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_5/index_2782_seed_13133_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 19:04:26.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:26.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:35.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_5/index_2782_seed_13134_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:05:55.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 19:16:46.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:52.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_5/index_9946_seed_28219_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 19:18:09.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:18:09.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:19.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_5/index_9946_seed_28220_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:19:40.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.331 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.807 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 19:31:09.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_5/index_1896_seed_21232_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 19:32:31.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:31.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:33:43.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_5/index_1896_seed_21233_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:34:07.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 19:45:03.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_5/index_1339_seed_8597_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 19:46:23.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:23.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:32.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_5/index_1339_seed_8598_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:47:52.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 19:58:45.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_5/index_4557_seed_14276_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 20:00:08.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:00:08.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:16.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_5/index_4557_seed_14277_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:01:38.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 20:12:28.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_5/index_9035_seed_21635_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 20:13:48.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:48.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:14:58.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_5/index_9035_seed_21636_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:15:22.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 20:26:14.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_5/index_560_seed_20670_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 20:27:36.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:36.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:28:48.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_5/index_560_seed_20671_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:29:12.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 20:40:06.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:06.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_5/index_5985_seed_13882_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 20:41:28.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:28.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:42:38.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_5/index_5985_seed_13883_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:43:01.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 20:53:52.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_5/index_5366_seed_25981_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 20:55:15.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:55:15.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:25.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_5/index_5366_seed_25982_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:56:54.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 21:07:44.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:50.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_5/index_2763_seed_5683_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 21:09:04.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:09:04.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:18.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_5/index_2763_seed_5684_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:10:42.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 21:21:31.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:37.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_5/index_7834_seed_30378_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 21:22:55.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:55.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:02.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_5/index_7834_seed_30379_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:24:26.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 21:35:16.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_5/index_1152_seed_31508_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 21:36:39.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:39.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:37:49.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_5/index_1152_seed_31509_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:38:11.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.291 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 21:49:43.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:51.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_5/index_9693_seed_7729_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 21:51:08.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:51:08.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:15.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_5/index_9693_seed_7730_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:52:38.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 22:03:26.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_5/index_945_seed_24625_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 22:04:49.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:49.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:05:55.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_5/index_945_seed_24626_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:06:18.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 22:17:07.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_5/index_1538_seed_22719_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 22:18:29.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:29.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:36.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_5/index_1538_seed_22720_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:19:59.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 22:30:49.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_5/index_9753_seed_14662_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 22:32:08.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:32:08.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:20.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_5/index_9753_seed_14663_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:33:43.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 22:44:32.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:38.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_5/index_8598_seed_10515_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 22:45:55.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:55.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:03.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_5/index_8598_seed_10516_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:47:24.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 22:59:16.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_5/index_5680_seed_11052_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 23:00:39.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:39.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:01:47.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_5/index_5680_seed_11053_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:02:09.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 23:13:02.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_5/index_2504_seed_18987_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 23:14:22.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:22.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:31.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_5/index_2504_seed_18988_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:15:53.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 23:26:44.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_5/index_2910_seed_1617_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 23:28:03.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:28:03.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:15.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_5/index_2910_seed_1618_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:29:39.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 23:40:27.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:33.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_5/index_5113_seed_18997_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 23:41:56.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:56.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:01.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_5/index_5113_seed_18998_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:43:26.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-26 23:54:16.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_5/index_6194_seed_19025_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-26 23:55:36.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:36.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:56:46.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_5/index_6194_seed_19026_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:57:09.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:39.118 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 00:08:40.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:48.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_5/index_1402_seed_4134_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 00:10:03.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:10:03.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:11.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_5/index_1402_seed_4135_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:11:32.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 00:22:20.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_5/index_3825_seed_9258_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 00:23:39.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:39.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:24:53.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_5/index_3825_seed_9259_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:25:15.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 00:36:06.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_5/index_7367_seed_23915_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 00:37:26.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:26.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:38.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_5/index_7367_seed_23916_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:38:59.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 00:49:49.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_5/index_2061_seed_3763_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 00:51:13.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:51:13.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:19.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_5/index_2061_seed_3764_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:52:45.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 01:03:35.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_5/index_1695_seed_9585_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 01:04:56.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:56.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:05.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_5/index_1695_seed_9586_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:06:28.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 01:17:19.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_5/index_2594_seed_15550_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 01:18:38.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:38.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:19:52.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_5/index_2594_seed_15551_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:20:15.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 01:31:06.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_5/index_366_seed_29061_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 01:32:26.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:26.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:38.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_5/index_366_seed_29062_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:33:58.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 01:44:47.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_5/index_650_seed_14521_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 01:46:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:46:06.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:19.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_5/index_650_seed_14522_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:47:40.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 01:58:27.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:33.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_5/index_3219_seed_17220_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 01:59:48.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:48.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:00.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_5/index_3219_seed_17221_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:01:23.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 02:12:14.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_5/index_1182_seed_5532_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 02:13:37.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:37.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:14:47.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_5/index_1182_seed_5533_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:15:09.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.300 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 02:26:39.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_5/index_2552_seed_8151_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 02:28:00.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:28:00.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:12.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_5/index_2552_seed_8152_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:29:33.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 02:40:23.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_5/index_8201_seed_19121_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 02:41:46.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:46.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:00.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_5/index_8201_seed_19122_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:43:25.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 02:54:14.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:20.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_5/index_5187_seed_16770_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 02:55:35.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:35.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:56:46.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_5/index_5187_seed_16771_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:57:09.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 03:08:01.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:02.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_5/index_8585_seed_3420_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 03:09:23.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:23.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:31.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_5/index_8585_seed_3421_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:10:50.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 03:21:41.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:47.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_5/index_8967_seed_25115_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 03:23:08.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:23:08.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:15.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_5/index_8967_seed_25116_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:24:38.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 03:35:26.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_5/index_8088_seed_5986_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 03:36:49.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:49.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:02.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_5/index_8088_seed_5987_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:38:25.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 03:49:14.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_5/index_5626_seed_7595_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 03:50:35.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:35.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:51:45.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_5/index_5626_seed_7596_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:52:08.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 04:03:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_5/index_5530_seed_11628_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 04:04:23.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:23.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:33.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_5/index_5530_seed_11629_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:05:54.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 04:16:48.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_5/index_375_seed_14694_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 04:18:07.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:18:07.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:19.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_5/index_375_seed_14695_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:19:42.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 04:30:31.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_5/index_6576_seed_7145_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 04:31:53.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:53.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:01.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_5/index_6576_seed_7146_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:33:22.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.481 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 04:45:58.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_5/index_2256_seed_7918_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 04:47:20.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:20.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:31.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_5/index_2256_seed_7919_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:48:56.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 04:59:46.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_5/index_5615_seed_19829_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 05:01:13.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:01:13.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:17.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_5/index_5615_seed_19830_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:02:40.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 05:13:27.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:27.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:32.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_5/index_3719_seed_30286_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 05:14:49.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:49.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:15:56.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_5/index_3719_seed_30287_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:16:17.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 05:27:07.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:08.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_5/index_6069_seed_18030_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 05:28:35.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:35.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:29:40.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_5/index_6069_seed_18031_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:30:01.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 05:40:50.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_5/index_8408_seed_6823_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 05:42:14.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:42:14.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:19.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_5/index_8408_seed_6824_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:43:44.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 05:54:35.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_5/index_7397_seed_7236_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 05:56:01.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:56:01.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:06.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_5/index_7397_seed_7237_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:57:30.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 06:08:19.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:25.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_5/index_9911_seed_28658_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 06:09:39.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:39.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:10:49.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_5/index_9911_seed_28659_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:11:14.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 06:22:07.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:07.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_5/index_616_seed_15305_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 06:23:32.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:32.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:38.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_5/index_616_seed_15306_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:24:59.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 06:35:50.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_5/index_8696_seed_18457_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 06:37:09.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:37:09.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:20.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_5/index_8696_seed_18458_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:38:42.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 06:49:30.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:36.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_5/index_410_seed_28830_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 06:50:51.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:51.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:02.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_5/index_410_seed_28831_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:52:21.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.842 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 07:03:49.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_5/index_3810_seed_11627_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 07:05:11.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:05:11.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:20.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_5/index_3810_seed_11628_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:06:40.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 07:17:29.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_5/index_4960_seed_25292_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 07:18:53.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:53.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:00.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_5/index_4960_seed_25293_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:20:22.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 07:31:16.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:22.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_5/index_3893_seed_17508_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 07:32:36.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:36.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:33:49.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_5/index_3893_seed_17509_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:34:12.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 07:45:07.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:13.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_5/index_364_seed_23848_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 07:46:26.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:26.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:47:37.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_5/index_364_seed_23849_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:48:02.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 07:58:52.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_5/index_4102_seed_31464_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 08:00:16.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:00:16.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:23.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_5/index_4102_seed_31465_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:01:46.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 08:12:38.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:42.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_5/index_8638_seed_31654_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 08:13:58.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:58.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:10.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_5/index_8638_seed_31655_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:15:31.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 08:26:20.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:26.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_5/index_6042_seed_32801_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 08:27:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:42.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:28:51.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_5/index_6042_seed_32802_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:29:11.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 08:40:02.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_5/index_3167_seed_16254_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 08:41:24.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:24.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:35.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_5/index_3167_seed_16255_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:42:57.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 08:53:47.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:54.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_5/index_3582_seed_27117_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 08:55:12.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:55:12.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:23.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_5/index_3582_seed_27118_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:56:50.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 09:07:41.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_5/index_5612_seed_33202_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 09:09:00.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:09:00.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:13.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_5/index_5612_seed_33203_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:10:37.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.069 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 09:22:02.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:02.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_5/index_6669_seed_7262_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 09:23:25.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:25.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_5/index_6669_seed_7263_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:24:55.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 09:35:43.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_5/index_6319_seed_4395_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 09:37:03.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:37:03.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:13.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_5/index_6319_seed_4396_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:38:35.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 09:49:26.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_5/index_5434_seed_31497_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 09:50:47.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:47.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:51:57.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_5/index_5434_seed_31498_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:52:18.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-27 09:54:10.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 10:03:08.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_5/index_6420_seed_28673_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 10:04:28.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:28.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:38.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_5/index_6420_seed_28674_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:06:00.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 10:16:53.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_5/index_8589_seed_30649_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 10:18:15.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:18:15.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:23.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_5/index_8589_seed_30650_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:19:46.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 10:31:49.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:54.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_5/index_1107_seed_23307_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 10:33:10.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:33:10.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:21.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_5/index_1107_seed_23308_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:34:43.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 10:45:31.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_5/index_1941_seed_14823_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 10:46:54.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:54.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:03.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_5/index_1941_seed_14824_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:48:27.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 10:59:16.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_5/index_3348_seed_2700_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 11:00:40.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:40.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:01:49.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_5/index_3348_seed_2701_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:02:13.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.909 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 11:13:07.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_5/index_5214_seed_13870_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 11:14:25.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:25.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:35.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_5/index_5214_seed_13871_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:15:55.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 11:26:47.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_5/index_6548_seed_29978_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 11:28:09.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:28:09.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:21.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_5/index_6548_seed_29979_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:29:42.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.484 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 11:41:06.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_5/index_4685_seed_6106_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 11:42:31.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:31.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:37.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_5/index_4685_seed_6107_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:43:59.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 11:54:48.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_5/index_7137_seed_14472_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 11:56:08.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:56:08.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:19.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_5/index_7137_seed_14473_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:57:45.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 12:08:34.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_5/index_3236_seed_19089_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 12:09:56.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:56.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:05.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_5/index_3236_seed_19090_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:11:27.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 12:22:19.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:25.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_5/index_9002_seed_7237_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 12:23:41.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:41.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:24:50.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_5/index_9002_seed_7238_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:25:10.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 12:36:01.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_5/index_4451_seed_21153_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 12:37:23.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:23.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:32.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_5/index_4451_seed_21154_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:38:53.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 12:49:44.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_5/index_192_seed_20093_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 12:51:09.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:51:09.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:14.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_5/index_192_seed_20094_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:52:34.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 13:03:21.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_5/index_2679_seed_3183_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 13:04:45.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:45.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:05:51.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_5/index_2679_seed_3184_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:06:13.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 13:17:07.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:12.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_5/index_2694_seed_21388_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 13:18:31.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:31.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:36.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_5/index_2694_seed_21389_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:19:58.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 13:30:47.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_5/index_1743_seed_4959_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 13:32:13.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:32:13.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:21.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_5/index_1743_seed_4960_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:33:47.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 13:44:38.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_5/index_6029_seed_16536_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 13:46:01.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:46:01.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:08.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_5/index_6029_seed_16537_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:47:31.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.306 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 13:58:57.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_5/index_666_seed_11654_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 14:00:18.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:18.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:30.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_5/index_666_seed_11655_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:01:51.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 14:12:43.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_5/index_4221_seed_15055_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 14:14:03.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:14:03.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:11.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_5/index_4221_seed_15056_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 14:15:31.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:15:32.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 14:26:20.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:26.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_5/index_3344_seed_11182_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 14:27:41.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:41.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:28:50.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_5/index_3344_seed_11183_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:29:12.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 14:40:05.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:06.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:10.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_5/index_6557_seed_23935_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 14:41:25.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:25.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:36.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_5/index_6557_seed_23936_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:42:58.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 14:53:47.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_5/index_2394_seed_20693_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 14:55:06.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:55:06.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:17.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_5/index_2394_seed_20694_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:56:39.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.324 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 15:07:26.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:31.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_5/index_9971_seed_13359_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 15:08:47.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:47.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:09:56.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_5/index_9971_seed_13360_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:10:18.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 15:21:06.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:12.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_5/index_1571_seed_10552_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 15:22:26.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:26.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:37.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_5/index_1571_seed_10553_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:23:58.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 15:34:48.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_5/index_6506_seed_22417_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 15:36:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:36:07.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:20.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_5/index_6506_seed_22418_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:37:44.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 15:48:32.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_5/index_213_seed_21858_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 15:49:53.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:03.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_5/index_213_seed_21859_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:51:25.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 16:02:14.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_5/index_2292_seed_23324_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 16:03:33.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:33.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:04:44.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_5/index_2292_seed_23325_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:05:05.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.464 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.937 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 16:17:43.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_5/index_9182_seed_26347_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 16:19:07.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:19:07.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:17.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_5/index_9182_seed_26348_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:20:37.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 16:31:26.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_5/index_7051_seed_16090_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 16:32:48.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:48.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:33:58.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_5/index_7051_seed_16091_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:34:18.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 16:45:07.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_5/index_4293_seed_18147_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 16:46:28.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:28.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:38.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_5/index_4293_seed_18148_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:47:59.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 16:58:47.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:53.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_5/index_9240_seed_9293_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 17:00:08.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:00:08.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:20.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_5/index_9240_seed_9294_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:01:41.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 17:12:31.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_5/index_1279_seed_29260_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 17:13:51.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:51.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:02.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_5/index_1279_seed_29261_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:15:22.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 17:26:15.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_5/index_4750_seed_28043_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 17:27:32.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:32.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:28:46.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_5/index_4750_seed_28044_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:29:09.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 17:40:00.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:00.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_5/index_9852_seed_16596_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 17:41:19.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:19.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:34.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_5/index_9852_seed_16597_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:42:55.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 17:53:45.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:51.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_5/index_2969_seed_28400_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 17:55:08.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:55:08.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:15.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_5/index_2969_seed_28401_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:56:38.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 18:07:25.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_5/index_4773_seed_26044_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 18:08:56.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:56.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:01.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_5/index_4773_seed_26045_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:10:24.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 18:21:14.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_5/index_1554_seed_26746_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 18:22:37.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:37.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:23:45.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_5/index_1554_seed_26747_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:24:07.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.370 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 18:35:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:44.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_5/index_1383_seed_240_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 18:37:00.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:37:00.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:08.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_5/index_1383_seed_241_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:38:32.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 18:49:21.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_5/index_687_seed_7734_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 18:50:45.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:45.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:51:55.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_5/index_687_seed_7735_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:52:16.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 19:03:06.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:12.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_5/index_2306_seed_8616_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 19:04:25.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:25.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:36.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_5/index_2306_seed_8617_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:05:55.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 19:16:45.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:50.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_5/index_8748_seed_25298_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 19:18:06.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:18:06.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:21.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_5/index_8748_seed_25299_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:19:42.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 19:30:30.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_5/index_6857_seed_33264_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 19:31:49.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:49.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:04.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_5/index_6857_seed_33265_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:33:25.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 19:44:14.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_5/index_4413_seed_16220_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 19:45:33.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:33.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:46:43.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_5/index_4413_seed_16221_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:47:10.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 19:58:02.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_5/index_3242_seed_5941_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 19:59:20.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:20.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:33.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_5/index_3242_seed_5942_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:00:54.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 20:11:43.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:43.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:48.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_5/index_5425_seed_4546_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 20:13:06.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:13:06.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:15.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_5/index_5425_seed_4547_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:14:39.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 20:25:26.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_5/index_2047_seed_879_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 20:26:45.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:45.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:27:58.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_5/index_2047_seed_880_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:28:19.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 20:39:06.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_5/index_5424_seed_16128_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 20:40:29.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:29.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:41:38.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_5/index_5424_seed_16129_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:42:04.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.552 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 20:53:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_5/index_8180_seed_19677_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 20:54:50.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:50.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:55:59.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_5/index_8180_seed_19678_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:56:21.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 21:07:10.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_5/index_7151_seed_26870_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 21:08:34.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:34.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:09:41.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_5/index_7151_seed_26871_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:10:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 21:21:01.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_5/index_1853_seed_8082_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 21:22:21.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:21.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:31.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_5/index_1853_seed_8083_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:23:52.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 21:34:42.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_5/index_7138_seed_21625_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 21:36:05.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:36:05.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:12.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_5/index_7138_seed_21626_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:37:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 21:48:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_5/index_2186_seed_4692_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 21:49:48.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:48.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:50:54.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_5/index_2186_seed_4693_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:51:16.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 22:03:21.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_5/index_8256_seed_30228_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 22:04:39.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:39.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:05:54.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_5/index_8256_seed_30229_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:06:17.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 22:17:06.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_5/index_4283_seed_28833_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 22:18:30.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:30.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:36.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_5/index_4283_seed_28834_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:19:57.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 22:30:47.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_5/index_9821_seed_17653_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 22:32:05.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:32:05.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:21.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_5/index_9821_seed_17654_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:33:41.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 22:44:30.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_5/index_1792_seed_30637_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 22:45:52.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:52.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:46:59.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_5/index_1792_seed_30638_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:47:18.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 22:58:07.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_5/index_4059_seed_573_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 22:59:29.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:29.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:00:37.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_5/index_4059_seed_574_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:01:01.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.385 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 23:12:29.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_5/index_4485_seed_13204_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 23:13:51.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:51.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:01.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_5/index_4485_seed_13205_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:15:23.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 23:26:15.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_5/index_9516_seed_18573_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 23:27:36.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:36.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:28:45.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_5/index_9516_seed_18574_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:29:06.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 23:40:02.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_5/index_4275_seed_21203_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 23:41:25.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:25.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:42:36.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_5/index_4275_seed_21204_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:43:01.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-27 23:53:51.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_5/index_7029_seed_15784_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-27 23:55:09.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:55:09.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:22.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_5/index_7029_seed_15785_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:56:44.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 00:07:32.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_5/index_1023_seed_9552_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 00:08:54.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:54.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:08.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_5/index_1023_seed_9553_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:10:31.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 00:21:21.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:21.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_5/index_4160_seed_33408_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 00:22:44.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:44.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:23:51.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_5/index_4160_seed_33409_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:24:12.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 00:35:07.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_5/index_6675_seed_15164_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 00:36:28.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:28.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:37:45.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_5/index_6675_seed_15165_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:38:10.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 00:49:02.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:08.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_5/index_1133_seed_4818_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 00:50:26.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:26.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:36.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_5/index_1133_seed_4819_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:51:57.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 01:02:48.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_5/index_5376_seed_26265_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 01:04:06.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:04:06.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:21.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_5/index_5376_seed_26266_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:05:44.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 01:16:33.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_5/index_4349_seed_11251_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 01:17:53.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:53.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:11.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_5/index_4349_seed_11252_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:19:31.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.799 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 01:30:58.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_5/index_7910_seed_31149_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 01:32:19.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:19.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:29.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_5/index_7910_seed_31150_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:33:52.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 01:44:44.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_5/index_115_seed_15774_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 01:46:03.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:46:03.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:15.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_5/index_115_seed_15775_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:47:37.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 01:58:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:33.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_5/index_5240_seed_31236_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 01:59:47.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:47.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:00:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_5/index_5240_seed_31237_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:01:18.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 02:12:08.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_5/index_9987_seed_30372_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 02:13:36.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:36.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:14:40.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_5/index_9987_seed_30373_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:15:03.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 02:25:53.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_5/index_3001_seed_30722_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 02:27:16.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:27:16.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:25.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_5/index_3001_seed_30723_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:28:45.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 02:39:38.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_5/index_4857_seed_11760_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 02:40:57.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:57.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:08.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_5/index_4857_seed_11761_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:42:31.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 02:53:20.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:21.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_5/index_9317_seed_8940_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 02:54:42.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:42.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:55:51.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_5/index_9317_seed_8941_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:56:13.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 03:07:08.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_5/index_8890_seed_3809_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 03:08:33.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:33.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:09:38.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_5/index_8890_seed_3810_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:10:03.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 03:20:53.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_5/index_8186_seed_7118_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 03:22:14.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:22:14.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:26.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_5/index_8186_seed_7119_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:23:49.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 03:34:40.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_5/index_8318_seed_24610_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 03:35:59.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:59.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:12.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_5/index_8318_seed_24611_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:37:35.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.521 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 03:50:13.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_5/index_4611_seed_15391_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 03:51:36.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:36.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:52:46.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_5/index_4611_seed_15392_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:53:06.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 04:04:03.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_5/index_5990_seed_22520_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 04:05:21.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:21.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:33.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_5/index_5990_seed_22521_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:06:56.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 04:17:48.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_5/index_7116_seed_5164_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 04:19:11.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:19:11.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:20.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_5/index_7116_seed_5165_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:20:47.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.283 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 04:31:40.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_5/index_9514_seed_27118_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 04:32:57.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:57.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:09.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_5/index_9514_seed_27119_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:34:29.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 04:45:20.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:26.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_5/index_6820_seed_15093_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 04:46:40.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:40.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:47:50.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_5/index_6820_seed_15094_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:48:10.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 04:59:03.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_5/index_6983_seed_19911_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 05:00:23.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:23.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:32.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_5/index_6983_seed_19912_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:01:52.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 05:12:44.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:49.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_5/index_1987_seed_13685_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 05:14:05.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:14:05.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:14.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_5/index_1987_seed_13686_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:15:34.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 05:26:26.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_5/index_5072_seed_2117_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 05:27:45.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:45.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:28:56.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_5/index_5072_seed_2118_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:29:22.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 05:40:15.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_5/index_791_seed_32018_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 05:41:38.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:38.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:42:47.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_5/index_791_seed_32019_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:43:10.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 05:54:06.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_5/index_5006_seed_20827_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 05:55:26.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:26.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:37.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_5/index_5006_seed_20828_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:57:00.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.864 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 06:08:27.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_5/index_4711_seed_17929_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 06:09:51.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:51.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:10:58.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_5/index_4711_seed_17930_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:11:21.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 06:22:11.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:16.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_5/index_7561_seed_15345_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 06:23:30.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:30.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:24:41.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_5/index_7561_seed_15346_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 06:25:06.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:25:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 06:36:02.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_5/index_4243_seed_24130_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 06:37:23.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:23.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:32.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_5/index_4243_seed_24131_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:38:53.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 06:49:45.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:51.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_5/index_6077_seed_19434_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 06:51:07.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:51:07.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:15.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_5/index_6077_seed_19435_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:52:37.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 07:03:27.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_5/index_17_seed_9884_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 07:04:49.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:49.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:05:59.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_5/index_17_seed_9885_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:06:21.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 07:17:12.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_5/index_759_seed_9021_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 07:18:30.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:30.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:19:42.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_5/index_759_seed_9022_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:20:04.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 07:30:53.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:54.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_5/index_122_seed_30384_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 07:32:13.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:32:13.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:26.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_5/index_122_seed_30385_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:33:48.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 07:44:40.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_5/index_6046_seed_6343_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 07:46:01.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:46:01.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:11.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_5/index_6046_seed_6344_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:47:33.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 07:58:21.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_5/index_3821_seed_13420_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 07:59:43.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:43.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:00:52.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_5/index_3821_seed_13421_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:01:16.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.262 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 08:12:08.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_5/index_116_seed_20086_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 08:13:30.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:30.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:14:38.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_5/index_116_seed_20087_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:15:03.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.638 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 08:26:31.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_5/index_2443_seed_644_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 08:27:52.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:52.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:02.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_5/index_2443_seed_645_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:29:27.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 08:40:22.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_5/index_1404_seed_11985_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 08:41:43.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:43.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:42:52.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_5/index_1404_seed_11986_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:43:14.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 08:54:07.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:07.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_5/index_4693_seed_22346_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 08:55:26.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:26.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:38.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_5/index_4693_seed_22347_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:57:00.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 09:07:50.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_5/index_4768_seed_8740_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 09:09:12.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:09:12.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:20.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_5/index_4768_seed_8741_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:10:44.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 09:21:32.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_5/index_5625_seed_22881_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 09:22:54.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:54.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:04.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_5/index_5625_seed_22882_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:24:28.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 09:36:35.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_5/index_8098_seed_2070_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 09:37:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:55.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:04.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_5/index_8098_seed_2071_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:39:27.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 09:50:22.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_5/index_8323_seed_480_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 09:51:43.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:43.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:52:55.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_5/index_8323_seed_481_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:53:25.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 10:04:16.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_5/index_9872_seed_17217_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 10:05:38.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:38.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:06:50.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_5/index_9872_seed_17218_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:07:11.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 10:18:09.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_5/index_5460_seed_26037_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 10:19:30.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:30.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:20:40.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_5/index_5460_seed_26038_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:21:02.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 10:31:54.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_5/index_3245_seed_19166_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 10:33:12.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:33:12.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:26.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_5/index_3245_seed_19167_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:34:54.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.869 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 10:46:20.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_5/index_9457_seed_6467_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 10:47:40.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:40.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:48:51.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_5/index_9457_seed_6468_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:49:13.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 11:00:08.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_5/index_9796_seed_11601_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 11:01:31.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:31.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:39.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_5/index_9796_seed_11602_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:03:00.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 11:13:51.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_5/index_4054_seed_33184_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 11:15:14.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:15:14.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:21.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_5/index_4054_seed_33185_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:16:43.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 11:27:32.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_5/index_1886_seed_17741_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 11:28:54.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:54.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:02.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_5/index_1886_seed_17742_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:30:23.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 11:41:15.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_5/index_5791_seed_16250_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 11:42:35.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:35.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:43:46.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_5/index_5791_seed_16251_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:44:08.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 11:55:03.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_5/index_4761_seed_18908_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 11:56:23.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:23.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:33.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_5/index_4761_seed_18909_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:57:57.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 12:08:48.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_5/index_428_seed_22031_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 12:10:12.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:10:12.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:25.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_5/index_428_seed_22032_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:11:48.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 12:22:40.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_5/index_9184_seed_20602_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 12:24:04.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:24:04.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:13.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_5/index_9184_seed_20603_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:25:34.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 12:36:24.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_5/index_9377_seed_16897_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 12:37:42.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:42.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:38:54.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_5/index_9377_seed_16898_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:39:17.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 12:50:08.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_5/index_9262_seed_18426_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 12:51:35.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:35.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:52:41.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_5/index_9262_seed_18427_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:53:04.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.914 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 13:04:33.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_5/index_9544_seed_22002_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 13:06:00.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:06:00.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:07.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_5/index_9544_seed_22003_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:07:33.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 13:18:22.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:27.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_5/index_68_seed_23086_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 13:19:42.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:42.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:20:53.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_5/index_68_seed_23087_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:21:14.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.395 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 13:32:07.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_5/index_4612_seed_15494_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 13:33:30.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:30.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:38.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_5/index_4612_seed_15495_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:35:00.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 13:45:49.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:54.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_5/index_7575_seed_17362_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 13:47:11.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:47:11.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:22.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_5/index_7575_seed_17363_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:48:42.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 13:59:30.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:30.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:35.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_5/index_7326_seed_21167_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 14:00:51.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:51.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:00.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_5/index_7326_seed_21168_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:02:21.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 14:13:08.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_5/index_7208_seed_18632_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 14:14:34.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:34.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:15:42.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_5/index_7208_seed_18633_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:16:09.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 14:27:02.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_5/index_3374_seed_21572_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 14:28:23.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:23.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:31.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_5/index_3374_seed_21573_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:29:51.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 14:40:44.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_5/index_7903_seed_25557_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 14:42:04.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:42:04.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:15.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_5/index_7903_seed_25558_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:43:36.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 14:54:25.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_5/index_9246_seed_16257_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 14:55:45.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:45.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:56:55.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_5/index_9246_seed_16258_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:57:15.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.484 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 15:08:07.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_5/index_50_seed_7486_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 15:09:29.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:29.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:10:38.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_5/index_50_seed_7487_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:11:02.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.307 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 15:23:36.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_5/index_9456_seed_25896_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 15:24:56.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:56.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:08.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_5/index_9456_seed_25897_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:26:32.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 15:37:21.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_5/index_2477_seed_22217_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 15:38:41.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:41.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:39:52.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_5/index_2477_seed_22218_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:40:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 15:51:08.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_5/index_973_seed_3781_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 15:52:30.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:30.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:53:40.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_5/index_973_seed_3782_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:54:04.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 16:04:54.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_5/index_2079_seed_20301_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 16:06:14.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:06:14.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:23.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_5/index_2079_seed_20302_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:07:50.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 16:18:40.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:41.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_5/index_2440_seed_2808_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 16:20:02.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:20:02.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:12.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_5/index_2440_seed_2809_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:21:34.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 16:32:23.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_5/index_7056_seed_32305_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 16:33:44.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:44.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:34:53.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_5/index_7056_seed_32306_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:35:14.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 16:46:07.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_5/index_3226_seed_17869_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 16:47:30.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:30.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:39.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_5/index_3226_seed_17870_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:49:00.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 16:59:50.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_5/index_3337_seed_20875_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 17:01:13.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:01:13.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:21.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_5/index_3337_seed_20876_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:02:44.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 17:13:34.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:40.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_5/index_7800_seed_21152_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 17:14:58.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:58.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:07.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_5/index_7800_seed_21153_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:16:31.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 17:27:21.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_5/index_6242_seed_25802_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 17:28:46.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:46.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:29:52.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_5/index_6242_seed_25803_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:30:19.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.798 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 17:41:44.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_5/index_5047_seed_11113_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 17:43:09.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:43:09.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:20.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_5/index_5047_seed_11114_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:44:43.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 17:55:31.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:32.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_5/index_6040_seed_7961_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 17:56:55.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:55.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:02.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_5/index_6040_seed_7962_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:58:25.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 18:09:15.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_5/index_1050_seed_32024_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 18:10:35.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:35.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:11:49.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_5/index_1050_seed_32025_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:12:10.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 18:23:03.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_5/index_1816_seed_21736_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 18:24:25.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:25.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:33.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_5/index_1816_seed_21737_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:25:54.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 18:36:44.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:49.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_5/index_5706_seed_22159_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 18:38:06.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:38:06.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:15.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_5/index_5706_seed_22160_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:39:39.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 18:50:26.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_5/index_6852_seed_20137_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 18:51:46.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:46.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:52:59.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_5/index_6852_seed_20138_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:53:22.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 19:04:13.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:14.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_5/index_4666_seed_6079_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 19:05:40.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:40.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:06:45.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_5/index_4666_seed_6080_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:07:08.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 19:18:02.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_5/index_6345_seed_909_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 19:19:23.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:23.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:30.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_5/index_6345_seed_910_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:20:53.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 19:31:44.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_5/index_6751_seed_16230_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 19:33:04.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:33:04.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:15.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_5/index_6751_seed_16231_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:34:37.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 19:45:24.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:24.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:29.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_5/index_3565_seed_31824_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 19:46:45.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:45.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:47:55.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_5/index_3565_seed_31825_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:48:19.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.384 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 19:59:44.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:50.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_5/index_3192_seed_20985_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 20:01:05.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:01:05.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:14.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_5/index_3192_seed_20986_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:02:35.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 20:13:25.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_5/index_6254_seed_6063_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 20:14:46.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:46.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:15:55.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_5/index_6254_seed_6064_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:16:16.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 20:27:07.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_5/index_6260_seed_11896_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 20:28:27.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:27.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:38.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_5/index_6260_seed_11897_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:29:59.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 20:40:49.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_5/index_4143_seed_4298_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 20:42:08.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:42:08.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:22.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_5/index_4143_seed_4299_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:43:46.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 20:54:36.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_5/index_2253_seed_23644_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 20:55:55.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:55.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:09.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_5/index_2253_seed_23645_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:57:34.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 21:09:41.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:46.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_5/index_8347_seed_5750_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 21:11:03.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:11:03.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:13.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_5/index_8347_seed_5751_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:12:35.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 21:23:24.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_5/index_9807_seed_12852_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 21:24:43.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:43.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:25:53.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_5/index_9807_seed_12853_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:26:15.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 21:37:07.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_5/index_2058_seed_29422_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 21:38:30.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:30.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_5/index_2058_seed_29423_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:39:58.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 21:50:48.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:53.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_5/index_750_seed_1555_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 21:52:09.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:52:09.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:21.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_5/index_750_seed_1556_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:53:43.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 22:04:31.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_5/index_8654_seed_11628_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 22:05:59.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:59.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:04.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_5/index_8654_seed_11629_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:07:26.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.719 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 22:18:53.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_5/index_388_seed_2126_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 22:20:15.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:20:15.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:24.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_5/index_388_seed_2127_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:21:47.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 22:32:40.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_5/index_8350_seed_19688_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 22:33:59.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:59.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:11.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_5/index_8350_seed_19689_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:35:35.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 22:46:24.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_5/index_3686_seed_12079_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 22:47:46.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:46.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:48:56.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_5/index_3686_seed_12080_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:49:18.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.312 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 23:00:07.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_5/index_7351_seed_17370_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 23:01:29.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:29.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:38.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_5/index_7351_seed_17371_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:02:59.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 23:13:49.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:55.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_5/index_1016_seed_4872_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 23:15:08.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:15:08.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:20.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_5/index_1016_seed_4873_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:16:44.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 23:27:32.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_5/index_1706_seed_7130_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 23:28:51.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:51.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:02.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_5/index_1706_seed_7131_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:30:23.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 23:41:14.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_5/index_1658_seed_20018_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 23:42:37.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:37.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:43:44.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_5/index_1658_seed_20019_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:44:06.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-28 23:55:00.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_5/index_5788_seed_13514_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-28 23:56:20.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:20.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:28.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_5/index_5788_seed_13515_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:57:50.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.806 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-29 00:08:39.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_5/index_8673_seed_6279_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-29 00:10:00.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:10:00.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:08.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_5/index_8673_seed_6280_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:11:29.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-29 00:22:19.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:25.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_5/index_8390_seed_8658_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-29 00:23:39.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:39.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:24:51.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_5/index_8390_seed_8659_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:25:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.189 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-29 00:36:43.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_5/index_3528_seed_2453_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-29 00:38:06.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:38:06.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:15.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_5/index_3528_seed_2454_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:39:39.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-29 00:50:27.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_5/index_3789_seed_1489_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-29 00:51:51.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:51.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:52:56.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_5/index_3789_seed_1490_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:53:19.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 22, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 22, dp_rank 11: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=5, rank_in_group=1 +2026-06-29 01:04:09.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_5/index_949_seed_33162_dp_11_rank_22_sample_2_slot_0.mp4 +2026-06-29 01:05:33.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 22 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:33.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 22 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:06:40.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 22: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_5/index_949_seed_33163_dp_11_rank_22_sample_3_slot_1.mp4 +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 22: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 01:07:03.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 22: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 22: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 22, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_25.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_25.log new file mode 100644 index 0000000000000000000000000000000000000000..898ee8034348189b0f1e6cb282ad2615da8c03c8 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_25.log @@ -0,0 +1,5147 @@ +2026-05-29 20:45:36.391 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:45:36.392 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 20:51:46.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:11.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:11.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 25, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 21:06:57.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:01.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 21:21:17.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:15.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 21:35:35.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:43.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 21:49:56.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:04.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.289 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 22:04:18.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 22:18:35.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:41.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 22:32:58.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:58.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:00.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 22:47:20.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:27.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:27.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.731 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 23:01:44.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:51.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:51.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:49.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:53.520 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 23:16:57.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 23:31:14.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 23:45:29.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:35.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:35.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-29 23:59:50.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:50.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:56.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:56.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 00:14:13.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 00:28:36.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:45.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 00:42:58.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:05.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:05.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 00:57:15.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:22.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 01:11:38.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.806 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 01:26:05.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.428 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.500 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 01:41:19.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:27.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:27.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:18.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 01:55:38.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:44.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:39.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 02:09:58.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:02.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 02:24:23.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:30.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:30.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 02:38:56.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:02.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:02.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:58.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 02:54:31.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:37.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:37.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 03:08:50.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:56.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:56.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:53.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 03:23:09.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:16.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:16.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 03:37:26.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.975 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 03:51:59.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:06.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.639 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 04:07:19.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 04:21:48.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:58.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 04:36:12.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 04:50:41.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 05:05:07.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:13.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:12.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 05:19:25.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 05:33:38.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:44.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:44.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:37.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 05:47:59.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:58.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 06:02:21.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 06:16:55.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 06:19:59.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:11.868 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 06:32:17.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:16.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 06:46:37.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:43.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:37.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 07:00:59.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:06.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:06.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:00.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 07:15:21.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:22.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:41.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 07:30:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.021 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 07:44:30.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:38.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 07:59:00.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:08.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 08:13:34.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:40.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:40.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 08:28:15.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:23.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:23.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:23.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 08:42:53.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:02.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:25.546 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 08:59:29.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:39.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.038 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:33.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 09:14:04.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:11.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:17.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 09:28:42.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:51.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 09:43:17.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:25.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.220 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 09:57:46.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:53.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 10:12:07.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:06.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 10:26:24.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:30.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:30.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 10:40:46.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:53.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:53.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 10:55:08.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:16.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:16.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 11:09:40.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:49.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.464 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:04.164 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 11:25:07.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:17.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:21.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 11:39:42.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:50.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 11:54:06.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 12:08:30.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:37.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:30.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.189 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 12:22:53.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:02.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 12:37:12.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:11.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 12:51:30.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:58.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 13:05:59.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:06.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:04.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 13:20:24.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:31.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:31.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:30.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 13:34:48.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.262 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:59.441 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 13:50:02.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:11.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 14:04:33.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:34.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 14:19:43.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:54.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:54.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 14:34:38.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 14:49:20.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:27.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 15:05:01.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:09.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:04.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 15:19:35.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 15:34:11.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:21.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 15:48:48.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:51.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 16:03:17.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:25.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:47.073 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 16:18:48.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:01.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 16:33:33.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 16:48:15.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:54.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 17:02:55.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 17:17:35.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:44.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 17:32:10.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 17:46:54.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:03.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.302 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 18:01:35.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 18:16:18.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 18:31:18.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:19.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:20.163 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 18:47:24.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:39.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 19:02:13.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:35.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 19:17:12.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:27.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 19:31:55.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 19:47:03.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:12.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:45.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 20:01:46.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 20:16:29.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 20:31:20.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:02.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 20:46:03.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:12.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 21:00:47.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:26.838 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 21:17:31.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:40.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:40.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:38.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 21:32:09.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:50:56.168 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:50:56.169 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 21:55:22.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:49.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:49.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:08.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 25, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 22:10:52.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:11:59.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 22:25:33.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:41.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:41.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 22:40:03.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:12.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:11.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 22:54:38.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:46.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:50.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:19.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 23:09:20.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:28.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 23:24:05.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:14.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 23:38:46.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:54.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:09.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-30 23:53:10.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:17.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:17.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:15.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 00:07:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:53.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:55.350 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 00:22:59.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 00:37:30.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:38.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:38.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 00:52:07.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:13.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:13.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:22.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 01:06:44.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:53.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 01:21:16.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:17.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 01:35:36.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:43.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:43.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 01:50:03.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:10.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:07.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 02:04:34.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:42.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:39.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 25, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-05-31 02:19:01.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:08.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:02.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:00:42.523 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:00:42.524 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 18:04:09.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:30.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:30.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 25, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 18:18:41.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:35.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 18:32:26.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.324 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 18:46:08.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:13.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:03.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 19:00:02.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:02.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 19:13:49.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 19:27:32.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:23.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 19:41:17.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:22.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 19:55:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:14.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 20:09:02.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.988 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 20:23:32.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 20:37:18.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:23.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:23.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 20:51:03.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:57.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.281 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 21:04:50.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:55.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:55.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 21:18:43.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:48.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 21:32:31.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 21:46:16.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.975 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 22:00:05.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:11.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 22:13:50.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:55.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:55.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 22:27:39.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:44.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:57.736 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 22:41:59.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.909 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 22:55:45.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:50.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 23:09:30.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:34.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:34.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 23:23:15.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:19.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:19.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:11.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 23:37:07.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:12.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:02.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-25 23:51:52.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 00:05:36.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 00:19:21.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:26.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 00:33:07.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 00:47:03.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:00.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:29.159 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 01:01:30.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:31.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 01:15:18.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:23.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:12.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 01:29:08.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:14.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:14.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 01:42:58.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 01:56:44.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 02:10:31.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 02:24:16.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.245 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 02:38:08.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:13.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 02:41:00.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 02:51:53.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:59.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 03:05:40.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:46.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.909 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 03:20:06.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 03:33:52.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:57.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:57.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 03:47:40.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:45.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 04:01:21.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 04:15:09.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 04:29:03.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 04:42:46.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 04:56:27.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:32.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:32.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 05:10:10.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:03.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 05:24:03.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.077 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 05:39:29.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:35.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 05:53:10.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:15.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:15.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 06:06:52.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 06:20:40.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:45.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 06:34:21.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:26.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:26.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 06:48:09.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:15.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:15.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 07:02:03.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:55.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 07:15:48.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 07:29:32.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:37.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:31.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 07:43:21.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:44.176 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 07:57:45.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:52.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 08:11:33.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 08:25:18.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:07.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 08:39:03.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:57.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.142 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 08:52:50.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:54.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:54.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.542 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 09:06:39.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 09:20:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:27.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:27.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 09:34:09.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:15.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:15.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 09:47:56.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:57.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:01.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:01.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:51.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 10:01:45.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:34.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:16:00.075 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 10:16:01.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:58.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 10:29:52.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:44.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 10:43:37.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:29.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 10:57:21.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 11:11:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:09.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:14.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 11:26:08.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:14.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:14.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.143 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 11:39:55.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 11:53:37.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:41.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:41.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:31.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 12:07:21.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:12.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 12:21:06.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:12.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:29.058 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 12:35:30.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:37.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:37.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 12:49:16.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:20.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:20.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.745 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:09.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.063 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 13:03:03.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:07.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:07.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:53.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 13:16:44.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:48.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:48.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 13:30:31.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 13:44:15.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 13:58:07.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 14:11:49.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:39.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 14:25:28.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:32.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:32.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:25.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 14:39:15.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:20.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:20.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:13.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.587 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 14:53:45.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 15:07:27.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 15:21:11.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:15.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:15.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 15:34:56.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 15:48:45.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:49.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:49.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 16:02:26.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:31.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:31.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:17.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 16:16:06.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:11.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:11.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:04.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 16:29:54.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:30:59.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:30:59.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 16:43:38.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.655 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 16:57:20.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:25.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:25.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:50.691 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 17:12:52.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:47.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 17:26:39.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:44.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 17:40:20.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 17:54:06.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 18:07:49.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 18:21:38.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:37.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 18:35:27.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:32.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:32.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 18:49:15.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:19.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:19.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:05.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 19:03:02.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:07.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 19:16:46.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:51.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:30.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.685 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 19:31:08.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:07.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 19:45:03.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 19:58:45.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:45.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:50.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:50.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:38.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 20:12:28.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:22.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 20:26:14.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 20:40:07.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:11.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:11.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 20:53:52.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:53.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:57.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:57.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:54.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 21:07:44.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 21:21:31.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 21:35:16.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:21.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:21.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:11.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.858 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 21:49:43.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 22:03:26.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:31.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:31.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:17.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 22:17:07.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 22:30:49.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 22:44:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 22:59:16.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 23:13:02.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:08.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:53.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 23:26:44.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 23:40:27.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:26.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-26 23:54:16.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:20.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:20.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.388 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 00:08:40.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 00:22:20.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 00:36:06.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:59.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 00:49:49.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 01:03:35.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:40.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 01:17:19.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:24.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:24.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 01:31:07.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:12.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:58.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 01:44:47.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:52.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:52.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 01:58:27.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 02:12:14.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.695 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 02:26:39.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:45.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:45.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:34.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 02:40:23.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:27.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:27.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 02:54:14.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:09.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 03:08:01.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:07.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:50.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 03:21:41.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:46.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:46.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 03:35:26.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 03:49:14.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 04:03:03.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:08.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 04:16:48.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:53.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 04:30:31.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:35.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:35.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.447 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 04:45:58.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:05.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 04:59:47.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:47.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:51.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:51.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:40.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 05:13:28.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:32.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:32.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 05:27:07.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:12.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:12.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 05:40:50.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 05:54:35.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:40.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 06:08:19.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:14.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 06:22:07.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:07.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:12.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:12.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 06:35:50.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:55.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 06:49:30.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:30.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.888 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 07:03:49.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:55.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:55.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 07:17:29.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:34.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:22.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 07:31:16.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:12.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 07:45:07.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 07:58:52.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:56.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:56.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 08:12:38.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:42.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:42.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 08:26:20.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:10.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 08:40:02.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:07.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:07.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 08:53:47.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:47.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:52.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 09:07:41.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:46.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.139 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 09:22:03.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 09:35:43.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:48.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 09:49:25.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:30.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:30.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:18.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 10:03:08.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:13.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.863 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 10:16:53.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:58.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 10:31:49.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:54.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:54.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 10:45:31.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:32.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 10:59:16.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:21.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:21.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:12.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 11:13:07.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:12.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 11:26:47.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:41.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.409 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 11:41:06.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:11.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:11.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.131 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:59.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 11:54:48.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:52.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:52.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 12:08:34.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:39.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 12:22:19.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 12:36:01.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 12:49:44.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 13:03:21.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 13:17:08.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:12.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:12.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 13:30:47.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:47.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 13:44:38.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:43.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.202 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 13:58:57.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:02.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 14:12:43.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 14:26:20.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:24.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:24.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:12.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 14:40:05.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:10.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:10.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:58.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 14:53:47.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 15:07:26.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:26.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:32.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 15:21:06.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:11.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:58.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 15:34:47.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:52.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:52.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.324 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 15:48:32.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:36.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:36.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 16:02:14.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:05.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.883 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 16:17:43.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:49.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:49.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 16:31:26.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 16:45:07.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:59.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 16:58:47.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:52.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:42.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 17:12:31.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 17:26:15.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:20.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:09.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 17:40:00.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:05.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 17:53:45.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:38.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 18:07:25.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:24.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 18:21:14.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:19.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:19.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:07.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.795 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 18:35:38.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:43.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 18:49:21.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 19:03:06.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:11.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 19:16:45.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:50.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:50.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 19:30:30.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:25.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 19:44:14.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:18.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:18.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:10.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 19:58:02.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:54.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 20:11:43.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:48.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:48.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:39.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 20:25:26.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:19.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 20:39:06.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:12.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.630 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 20:53:29.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:34.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:21.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 21:07:10.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:15.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:07.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 21:21:01.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:52.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 21:34:42.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 21:48:23.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:23.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 22:03:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:16.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 22:17:06.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:57.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 22:30:47.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:52.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 22:44:30.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 22:58:07.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:11.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:12.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.784 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 23:12:29.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:35.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:35.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 23:26:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:20.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 23:40:02.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.70s (1.86min) +2026-06-27 23:44:54.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:50.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-27 23:53:51.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:55.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:55.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 00:07:32.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 00:21:21.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:21.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:27.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 00:35:07.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 00:49:02.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:07.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:57.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 01:02:48.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 01:16:33.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.644 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 01:30:58.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 01:44:44.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 01:58:27.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 02:12:08.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:14.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:14.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:04.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 02:25:53.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 02:39:38.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:31.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 02:53:20.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:25.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:25.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 03:07:08.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:03.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 03:20:53.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:49.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 03:34:40.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:45.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.774 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 03:50:13.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 04:04:03.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 04:17:48.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:53.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:53.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 04:31:40.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 04:45:20.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:10.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 04:59:03.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 05:12:44.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:49.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:49.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:34.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 05:26:25.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:21.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.246 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 05:40:15.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 05:54:06.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:11.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:11.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.927 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 06:08:27.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:21.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 06:22:11.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:16.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:16.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:06.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 06:36:02.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:07.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 06:49:45.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 07:03:27.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 07:17:12.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:17.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:04.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 07:30:53.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:48.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 07:44:40.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.045 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:33.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 07:58:21.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:26.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:26.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 08:12:08.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:14.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:14.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.743 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 08:26:31.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 08:40:22.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:27.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:27.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:14.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 08:54:07.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:13.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 08:57:00.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 09:07:50.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:51.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:56.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 09:21:32.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:37.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 09:36:35.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 09:50:22.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:25.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 10:04:16.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 10:18:09.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:14.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:02.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 10:31:54.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:53.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.825 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 10:46:20.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 11:00:08.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 11:13:51.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 11:27:32.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 11:41:15.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 11:55:03.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 12:08:48.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:53.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:48.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 12:22:40.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 12:36:24.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 12:50:08.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:09.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.842 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 13:04:33.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:33.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 13:18:22.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:27.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:27.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 13:32:07.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 13:45:49.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:49.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:55.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 13:59:30.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:35.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:35.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:21.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 14:13:08.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 14:27:02.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 14:40:43.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:48.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 14:54:25.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 15:08:07.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:02.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.163 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 15:23:36.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:32.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 15:37:21.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 15:51:08.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 16:04:54.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:50.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 16:18:40.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:45.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:45.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 16:32:23.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:28.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.431 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 16:46:07.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 16:59:50.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 17:13:34.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.289 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 17:27:21.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:26.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.775 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 17:41:44.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:49.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:49.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 17:55:31.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:32.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 18:09:15.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 18:23:03.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 18:36:44.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:44.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:49.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:49.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:39.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.245 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.523 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 18:50:26.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:22.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 19:04:13.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:19.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:08.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 19:18:02.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.252 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 19:31:44.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 19:45:24.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:24.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:30.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.379 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 19:59:44.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 20:13:24.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:24.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:29.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:29.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:15.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 20:27:07.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:12.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 20:40:49.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:53.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:53.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 20:54:36.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 21:09:41.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:46.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:46.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 21:23:23.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:29.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 21:37:07.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:48.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 21:50:48.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:54.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 22:04:32.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:37.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.467 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 22:18:53.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 22:32:40.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:35.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 22:46:24.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:29.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 23:00:07.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:59.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 23:13:49.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:54.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 23:27:32.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:23.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 23:41:14.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-28 23:55:00.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-29 00:08:39.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:29.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-29 00:22:19.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.405 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-29 00:36:43.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:48.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-29 00:50:27.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 25, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 25, dp_rank 12: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=6, rank_in_group=0 +2026-06-29 01:04:09.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 25 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:14.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 25 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 25: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:03.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 25: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 25: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 25, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_33.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_33.log new file mode 100644 index 0000000000000000000000000000000000000000..7d3786aeed18994e62205c29729dce3c87d49dab --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_33.log @@ -0,0 +1,5147 @@ +2026-05-29 20:45:20.417 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:45:20.418 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 20:51:46.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:11.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:11.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 33, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 21:06:57.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:01.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.439 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 21:21:17.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:15.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 21:35:35.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:43.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 21:49:56.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:03.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:03.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 22:04:18.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 22:18:35.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:41.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 22:32:58.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:00.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 22:47:20.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:27.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:27.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 23:01:44.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:51.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:51.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:49.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:52.901 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 23:16:57.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:57.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 23:31:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 23:45:29.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:35.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:35.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-29 23:59:50.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:56.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:56.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 00:14:13.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 00:28:36.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:45.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 00:42:58.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:05.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:05.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 00:57:15.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:21.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:21.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.484 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 01:11:38.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.923 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 01:26:05.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.745 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 01:41:19.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:27.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:27.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:18.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 01:55:38.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:44.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:39.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 02:09:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:02.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 02:24:23.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:30.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:30.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 02:38:56.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:03.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:57.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 02:54:31.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:37.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:37.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.214 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 03:08:50.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:56.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:56.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:53.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 03:23:09.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:16.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:16.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.442 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 03:37:26.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 03:51:59.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:06.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.398 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 04:07:19.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 04:21:48.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:58.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 04:36:12.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 04:50:41.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.265 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 05:05:07.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:13.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:12.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 05:19:25.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 05:33:38.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:44.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:44.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:37.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 05:47:59.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:58.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 06:02:21.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 06:16:55.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 06:19:59.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:11.961 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 06:32:17.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:16.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 06:46:36.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:43.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 07:00:59.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:00.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 07:15:21.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:22.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:41.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 07:30:08.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 07:44:30.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:37.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:37.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 07:59:00.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.901 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:08.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 08:13:33.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:40.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:40.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 08:28:15.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:23.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:23.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:23.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 08:42:53.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:02.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:24.663 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 08:59:29.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:39.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:33.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 09:14:05.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:11.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:17.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 09:28:42.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:50.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:50.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 09:43:17.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:24.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 09:57:46.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:53.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 10:12:07.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.855 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:06.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 10:26:24.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:30.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:30.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 10:40:46.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:53.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:53.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 10:55:08.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:16.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:16.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 11:09:40.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:49.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.201 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:02.719 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 11:25:07.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:17.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.076 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:20.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 11:39:42.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:50.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 11:54:06.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 12:08:30.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:36.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:36.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:29.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 12:22:53.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:01.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:01.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 12:37:12.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.052 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:10.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 12:51:30.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 13:05:59.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:06.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:04.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 13:20:24.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:31.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:31.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 13:34:48.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.471 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 13:50:02.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:11.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.875 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 14:04:33.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 14:19:43.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:54.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:54.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 14:34:38.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 14:49:19.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:27.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 15:05:01.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:09.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:04.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 15:19:35.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 15:34:11.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:20.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:20.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 15:48:48.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:51.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 16:03:17.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:25.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.668 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 16:18:48.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:01.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 16:33:33.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 16:48:15.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 17:02:55.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 17:17:35.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:44.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 17:32:10.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 17:46:54.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:02.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:02.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 18:01:35.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 18:16:18.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 18:31:18.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.308 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:18.127 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 18:47:24.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:39.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 19:02:13.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:34.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 19:17:12.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:26.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:53.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 19:31:55.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 19:47:03.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:12.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 20:01:46.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 20:16:29.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.572 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 20:31:20.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 20:46:03.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:11.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:11.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 21:00:47.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.455 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 21:17:31.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:40.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:40.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:37.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 21:32:09.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:50:48.043 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:50:48.043 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 21:55:22.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:47.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:47.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:08.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 33, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 22:10:52.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:53.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:11:59.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 22:25:33.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:41.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:41.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 22:40:03.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:04.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:11.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:11.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:10.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 22:54:38.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:38.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:45.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:45.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:50.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 23:09:20.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:27.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:27.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 23:24:05.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:14.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 23:38:46.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:53.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:53.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-30 23:53:10.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:17.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:17.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:15.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 00:07:41.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:52.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:52.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:53.930 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 00:22:59.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 00:37:30.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:38.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:38.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 00:52:07.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:13.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:13.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:22.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 01:06:44.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:52.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:52.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 01:21:16.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:17.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 01:35:36.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:43.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:43.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 01:50:03.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:10.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:06.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 02:04:34.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:42.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.764 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:39.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 33, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-05-31 02:19:01.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:07.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:07.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:02.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:00:36.471 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:00:36.474 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 18:04:09.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:30.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:30.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 33, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 18:18:41.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:34.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 18:32:26.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 18:46:08.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:13.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:03.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 19:00:02.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:02.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 19:13:49.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 19:27:32.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:23.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 19:41:17.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:21.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:21.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 19:55:08.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:13.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:13.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 20:09:02.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.727 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.917 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 20:23:32.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 20:37:18.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:19.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:24.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 20:51:03.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:56.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 21:04:50.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:55.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:55.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 21:18:43.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:48.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 21:32:31.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.628 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 21:46:16.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 22:00:05.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:10.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:11.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 22:13:50.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:55.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:55.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 22:27:39.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:44.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:58.105 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 22:42:00.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.063 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 22:55:45.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:50.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 23:09:30.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:35.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 23:23:15.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:20.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:10.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 23:37:07.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:12.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:02.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-25 23:51:52.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 00:05:36.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 00:19:21.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:25.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:25.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 00:33:07.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 00:47:03.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:00.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:28.841 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 01:01:30.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 01:15:17.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:17.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:22.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:22.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:12.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 01:29:08.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:13.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:13.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 01:42:58.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 01:56:44.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.165 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 02:10:31.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 02:24:16.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.246 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 02:38:08.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:12.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:12.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 02:40:59.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 02:51:53.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:54.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:59.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 03:05:40.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:41.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:45.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:45.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.611 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 03:20:06.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 03:33:52.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:57.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:57.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.186 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 03:47:40.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:44.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:44.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 04:01:21.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 04:15:09.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 04:29:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 04:42:46.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 04:56:27.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:33.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 05:10:10.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:03.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 05:24:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.218 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 05:39:29.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:35.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 05:53:10.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:15.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:15.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 06:06:52.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 06:20:39.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:45.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 06:34:21.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:22.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:26.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:26.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.499 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 06:48:09.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:14.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 07:02:03.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.348 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:54.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 07:15:48.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 07:29:32.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:36.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:36.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:30.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 07:43:21.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:44.218 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 07:57:45.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:45.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:53.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:53.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 08:11:33.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 08:25:18.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:07.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 08:39:03.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:56.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 08:52:50.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:54.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:54.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 09:06:39.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 09:20:22.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:27.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:27.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 09:34:09.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:14.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.159 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 09:47:56.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:01.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:01.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.361 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:51.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 10:01:45.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:34.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:16:00.251 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 10:16:01.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:58.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 10:29:52.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:44.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 10:43:37.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.281 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:29.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 10:57:21.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 11:11:08.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:09.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:13.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:13.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 11:26:08.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:13.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 11:39:55.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 11:53:37.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:41.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:41.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:30.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 12:07:21.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.286 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:11.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 12:21:06.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:07.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:12.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:29.112 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 12:35:30.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:31.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:37.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:37.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 12:49:16.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:21.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:09.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 13:03:03.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:07.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:07.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:52.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 13:16:44.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:48.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:48.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 13:30:31.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 13:44:15.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 13:58:07.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 14:11:49.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:38.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 14:25:28.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:32.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:32.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:24.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 14:39:15.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:21.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:12.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.717 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 14:53:45.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 15:07:27.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 15:21:10.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:16.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.111 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 15:34:56.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 15:48:45.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:50.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 16:02:26.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:31.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:31.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.921 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:17.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 16:16:06.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:11.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:11.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:03.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 16:29:54.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:30:59.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:30:59.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 16:43:38.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 16:57:20.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:25.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:25.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:50.960 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 17:12:52.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:47.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 17:26:39.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:44.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 17:40:20.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 17:54:06.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 18:07:49.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.980 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 18:21:38.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:36.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 18:35:27.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:32.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:32.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 18:49:15.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:19.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:19.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:05.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.77s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 19:03:02.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:06.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:06.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 19:16:46.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:52.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:52.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.878 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 19:31:08.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:06.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 19:45:03.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 19:58:46.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:51.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:37.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 20:12:28.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.128 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:21.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 20:26:14.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 20:40:07.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:11.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:11.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.483 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 20:53:52.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:53.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:58.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.176 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:53.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 21:07:44.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.549 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 21:21:31.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 21:35:16.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:16.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:21.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:21.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:10.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.317 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 21:49:43.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:43.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 22:03:26.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:27.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:31.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:32.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:17.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 22:17:07.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 22:30:49.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 22:44:32.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 22:59:16.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 23:13:02.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:07.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:07.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:52.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 23:26:44.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 23:40:27.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:26.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-26 23:54:16.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:21.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.588 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 00:08:40.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 00:22:20.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.490 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 00:36:06.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:58.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 00:49:49.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 01:03:35.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:40.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.584 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 01:17:19.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:24.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:24.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 01:31:07.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:11.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:11.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:58.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 01:44:47.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:53.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 01:58:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 02:12:14.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.620 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 02:26:39.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:45.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:33.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 02:40:23.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:27.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:27.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 02:54:14.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:08.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 03:08:01.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:07.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.205 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:49.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 03:21:41.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:46.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:46.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 03:35:26.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 03:49:14.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 04:03:03.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:08.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 04:16:48.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:53.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 04:30:31.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:35.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:35.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.157 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 04:45:58.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:04.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:04.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 04:59:46.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:52.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:39.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 05:13:27.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:32.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:32.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 05:27:07.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:12.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:12.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.549 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 05:40:50.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 05:54:35.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:39.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:39.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 06:08:19.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:13.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 06:22:07.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:07.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:12.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:12.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 06:35:50.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:54.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:54.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 06:49:30.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:30.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.916 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.964 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 07:03:49.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:55.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:55.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 07:17:29.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:34.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:22.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 07:31:16.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:11.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 07:45:07.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 07:58:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:56.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 08:12:38.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:42.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:42.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 08:26:20.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:10.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 08:40:02.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:07.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:07.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 08:53:47.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:47.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:53.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.418 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 09:07:41.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:46.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:00.732 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 09:22:02.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 09:35:43.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:48.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 09:49:25.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:30.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:30.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 10:03:08.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:12.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:12.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.262 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 10:16:53.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:57.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:57.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.587 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 10:31:49.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:49.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:54.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:54.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 10:45:31.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 10:59:16.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:21.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:21.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:12.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 11:13:07.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:12.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.439 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 11:26:47.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.798 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:41.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.538 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 11:41:06.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:11.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:11.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:58.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 11:54:48.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:53.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 12:08:34.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:39.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 12:22:19.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 12:36:01.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 12:49:44.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 13:03:21.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 13:17:07.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:07.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:12.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:12.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 13:30:47.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:47.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 13:44:38.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:42.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:42.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.454 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 13:58:57.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:02.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 14:12:43.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 14:26:20.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:24.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:25.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:12.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 14:40:05.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:10.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:10.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:57.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 14:53:47.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 15:07:26.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:31.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:31.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 15:21:06.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:11.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:57.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 15:34:47.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:47.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:53.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 15:48:32.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:37.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 16:02:14.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:04.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.908 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 16:17:43.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:44.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:49.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:49.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 16:31:26.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 16:45:07.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:58.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 16:58:47.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:53.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:53.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:41.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 17:12:31.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 17:26:15.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:19.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:19.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:08.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 17:40:00.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:00.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:05.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 17:53:45.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:37.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 18:07:25.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:23.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 18:21:14.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:19.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:19.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:07.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.374 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 18:35:38.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:43.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 18:49:21.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 19:03:06.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:11.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 19:16:46.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:51.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 19:30:30.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:24.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 19:44:14.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:18.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:18.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:10.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 19:58:02.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 20:11:43.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:49.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:38.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 20:25:26.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:18.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 20:39:06.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:11.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:11.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.373 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 20:53:29.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:34.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:20.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 21:07:10.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:15.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:06.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 21:21:01.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:51.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 21:34:42.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 21:48:23.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:23.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 22:03:21.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:16.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 22:17:06.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.152 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:56.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 22:30:47.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:53.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:53.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 22:44:30.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 22:58:07.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:11.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:11.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.848 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 23:12:29.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:35.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:35.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 23:26:15.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:19.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:19.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 23:40:02.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:50.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-27 23:53:51.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:56.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:32.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 00:07:32.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:33.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 00:21:21.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:26.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:26.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 00:35:07.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 00:49:02.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:07.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:56.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 01:02:48.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 01:16:33.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.709 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 01:30:58.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 01:44:44.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 01:58:27.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 02:12:08.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:13.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:03.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 02:25:53.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 02:39:38.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:30.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 02:53:20.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:25.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:25.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 03:07:08.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:02.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 03:20:53.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:48.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 03:34:40.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:44.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:44.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.480 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.708 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 03:50:13.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 04:04:03.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.563 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 04:17:48.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:53.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:53.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 04:31:40.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 04:45:20.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:10.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 04:59:03.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 05:12:44.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:49.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:49.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:34.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 05:26:26.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:21.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 05:40:15.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 05:54:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:12.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:25.174 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 06:08:27.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:21.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:11.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 06:22:11.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:16.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:16.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:06.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 06:36:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:07.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 06:49:45.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.807 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 07:03:27.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 07:17:12.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:17.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:03.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 07:30:53.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:54.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:47.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 07:44:40.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:32.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 07:58:21.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:26.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:26.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 08:12:08.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:13.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.350 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 08:26:31.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 08:40:22.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:28.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:13.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 08:54:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:13.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 08:56:59.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 09:07:50.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:51.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:55.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:55.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 09:21:32.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:37.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 09:36:35.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 09:50:22.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:24.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 10:04:16.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 10:18:09.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:14.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:02.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 10:31:54.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:53.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:19.035 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 10:46:20.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 11:00:08.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 11:13:51.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 11:27:32.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 11:41:15.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 11:55:03.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 12:08:48.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:53.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:47.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 12:22:40.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 12:36:24.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 12:50:08.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:09.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:13.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:13.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.797 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 13:04:33.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:33.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 13:18:22.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:27.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:27.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:13.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 13:32:07.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.611 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 13:45:50.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:54.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:54.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.544 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 13:59:30.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:30.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:35.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:35.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:21.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 14:13:08.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 14:27:02.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 14:40:44.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:49.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 14:54:25.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 15:08:07.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:11.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:12.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:01.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.018 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 15:23:36.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:31.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 15:37:21.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 15:51:08.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 16:04:54.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:49.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 16:18:40.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:46.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 16:32:23.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:27.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:27.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 16:46:07.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 16:59:50.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 17:13:34.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 17:27:21.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:25.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:25.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.835 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 17:41:44.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:49.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:49.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 17:55:31.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 18:09:15.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:15.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:20.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:20.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 18:23:03.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 18:36:44.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:49.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:49.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:38.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 18:50:26.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:21.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 19:04:13.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:18.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:18.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.109 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:07.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 19:18:02.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:06.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:06.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 19:31:44.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.662 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 19:45:24.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:29.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:29.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.517 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 19:59:44.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 20:13:24.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:24.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:29.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:29.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:15.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 20:27:07.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:12.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 20:40:49.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:53.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:53.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 20:54:36.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 21:09:41.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:46.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:46.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 21:23:24.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:28.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:28.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.845 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 21:37:07.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 21:50:48.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:53.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:53.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 22:04:31.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:36.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:36.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.694 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 22:18:53.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.144 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 22:32:40.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:35.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 22:46:24.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:28.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:28.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:06.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 23:00:07.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.268 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:59.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 23:13:49.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:54.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 23:27:32.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:22.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 23:41:14.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-28 23:55:00.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:04.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-29 00:08:39.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:28.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-29 00:22:19.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.086 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-29 00:36:43.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:48.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.559 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-29 00:50:27.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.756 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.326 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 33, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 33, dp_rank 16: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=0 +2026-06-29 01:04:09.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 33 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:14.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 33 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 33: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:03.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 33: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 33: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 33, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_34.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_34.log new file mode 100644 index 0000000000000000000000000000000000000000..2ce19e84c82a5bb329775264d086691534620ac0 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_34.log @@ -0,0 +1,6079 @@ +2026-05-29 20:45:40.778 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-29 20:45:40.781 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-29 20:45:41.460 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:45:41.460 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 20:51:46.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_8/index_1462_seed_13996_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 20:53:29.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:29.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:38.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_8/index_1462_seed_13997_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 20:55:17.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 34, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 21:06:57.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_8/index_5307_seed_18347_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 21:08:18.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:18.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:09:39.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_8/index_5307_seed_18348_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:10:01.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 21:21:17.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:25.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_8/index_1986_seed_27807_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 21:22:40.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:40.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:23:53.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_8/index_1986_seed_27808_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:24:15.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 21:35:35.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_8/index_159_seed_7234_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 21:37:01.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:37:02.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:14.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_8/index_159_seed_7235_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:38:36.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 21:49:56.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_8/index_4344_seed_4086_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 21:51:19.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:19.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:52:35.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_8/index_4344_seed_4087_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:53:01.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.129 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 22:04:18.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_8/index_9171_seed_24571_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 22:05:44.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:44.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:06:52.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_8/index_9171_seed_24572_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:07:15.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 22:18:35.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:42.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_8/index_3529_seed_31694_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 22:19:59.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:59.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:14.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_8/index_3529_seed_31695_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:21:38.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 22:32:58.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:58.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_8/index_1919_seed_23596_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 22:34:19.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:19.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:35.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_8/index_1919_seed_23597_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:36:00.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 22:47:20.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_8/index_7272_seed_13531_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 22:48:42.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:42.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:49:56.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_8/index_7272_seed_13532_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:50:20.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 23:01:44.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_8/index_9288_seed_24729_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 23:03:06.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:03:06.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:23.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_8/index_9288_seed_24730_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:04:49.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:53.167 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 23:16:57.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_8/index_1534_seed_7620_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 23:18:24.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:24.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:35.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_8/index_1534_seed_7621_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:19:57.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 23:31:14.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:21.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_8/index_6800_seed_18243_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 23:32:35.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:35.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:33:52.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_8/index_6800_seed_18244_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:34:15.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 23:45:29.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_8/index_2326_seed_13461_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-29 23:46:51.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:51.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:08.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_8/index_2326_seed_13462_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:48:31.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-29 23:59:50.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_8/index_5219_seed_26993_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 00:01:10.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:01:10.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:26.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_8/index_5219_seed_26994_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:02:55.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 00:14:13.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_8/index_7843_seed_21863_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 00:15:37.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:37.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:16:51.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_8/index_7843_seed_21864_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:17:18.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 00:28:36.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_8/index_296_seed_8939_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 00:30:04.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:30:04.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:16.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_8/index_296_seed_8940_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:31:39.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 00:42:58.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_8/index_7634_seed_15958_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 00:44:20.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:20.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:45:37.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_8/index_7634_seed_15959_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:46:02.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 00:57:15.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_8/index_6947_seed_26233_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 00:58:36.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:36.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:59:51.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_8/index_6947_seed_26234_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:00:16.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 01:11:38.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_8/index_3933_seed_7576_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 01:12:59.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:59.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:18.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_8/index_3933_seed_7577_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 01:14:42.923 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:14:44.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 01:26:05.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_8/index_9019_seed_28616_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 01:27:30.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:30.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:28:42.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_8/index_9019_seed_28617_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:29:04.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.354 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 01:41:19.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_8/index_2759_seed_21861_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 01:42:44.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:44.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:43:56.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_8/index_2759_seed_21862_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:44:18.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 01:55:38.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_8/index_8546_seed_29910_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 01:56:59.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:59.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_8/index_8546_seed_29911_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 01:58:38.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:58:39.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 02:09:58.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_8/index_3707_seed_28676_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 02:11:20.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:20.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:12:39.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_8/index_3707_seed_28677_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:13:02.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 02:24:23.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_8/index_7427_seed_6728_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 02:25:47.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:47.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:05.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_8/index_7427_seed_6729_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:27:33.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 02:38:56.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_8/index_6551_seed_26444_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 02:40:19.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:19.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:35.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_8/index_6551_seed_26445_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:41:58.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 02:54:31.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_8/index_891_seed_33474_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 02:55:52.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:52.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:08.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_8/index_891_seed_33475_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:57:31.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 03:08:50.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_8/index_5807_seed_1832_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 03:10:11.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:10:11.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:26.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_8/index_5807_seed_1833_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:11:53.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 03:23:09.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:17.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_8/index_792_seed_22653_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 03:24:37.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:37.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:25:46.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_8/index_792_seed_22654_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 03:26:07.442 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:26:08.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 03:37:26.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_8/index_6045_seed_8793_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 03:38:49.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:49.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:07.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_8/index_6045_seed_8794_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 03:40:38.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:40:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 03:51:59.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_8/index_6794_seed_11833_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 03:53:21.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:21.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:54:38.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_8/index_6794_seed_11834_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:55:04.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.284 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 04:07:19.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:28.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_8/index_1880_seed_13841_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 04:08:44.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:44.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:01.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_8/index_1880_seed_13842_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:10:28.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 04:21:48.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_8/index_5534_seed_15000_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 04:23:15.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:23:15.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:28.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_8/index_5534_seed_15001_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:24:54.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 04:36:12.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_8/index_5095_seed_19288_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 04:37:37.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:37.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:38:50.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_8/index_5095_seed_19289_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:39:16.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 04:50:41.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:48.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_8/index_8455_seed_992_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 04:52:08.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:52:08.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:19.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_8/index_8455_seed_993_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 04:53:43.265 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:53:44.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 05:05:07.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_8/index_1388_seed_22577_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 05:06:28.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:28.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:07:43.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_8/index_1388_seed_22578_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 05:08:12.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:08:12.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 05:19:25.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:32.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_8/index_4030_seed_29169_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 05:20:50.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:50.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:21:58.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_8/index_4030_seed_29170_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:22:20.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 05:33:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:45.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_8/index_9229_seed_2480_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 05:34:59.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:59.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:15.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_8/index_9229_seed_2481_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:36:37.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 05:47:59.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_8/index_9708_seed_3187_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 05:49:24.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:25.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:35.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_8/index_9708_seed_3188_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:50:58.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 06:02:21.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_8/index_1904_seed_24828_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 06:03:46.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:46.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:04.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_8/index_1904_seed_24829_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:05:33.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 06:16:55.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_8/index_5748_seed_21478_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 06:18:20.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:20.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:31.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_8/index_5748_seed_21479_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 06:19:58.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:19:59.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:11.943 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 06:32:17.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_8/index_1943_seed_27500_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 06:33:38.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:38.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:34:55.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_8/index_1943_seed_27501_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:35:16.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 06:46:36.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_8/index_6584_seed_12338_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 06:48:02.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:48:02.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:14.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_8/index_6584_seed_12339_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:49:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 07:00:59.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_8/index_815_seed_18789_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 07:02:20.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:20.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:38.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_8/index_815_seed_18790_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:04:00.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 07:15:21.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_8/index_8934_seed_2551_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 07:16:44.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:44.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:06.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_8/index_8934_seed_2552_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:18:42.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 07:30:08.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_8/index_4320_seed_24277_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 07:31:36.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:36.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:32:48.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_8/index_4320_seed_24278_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 07:33:10.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:33:11.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 07:44:30.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_8/index_4988_seed_31733_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 07:45:51.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:51.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:04.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_8/index_4988_seed_31734_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:47:30.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 07:59:00.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_8/index_402_seed_29366_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 08:00:23.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:24.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:01:41.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_8/index_402_seed_29367_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 08:02:07.901 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:02:08.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 08:13:34.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_8/index_7811_seed_14301_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 08:15:02.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:15:02.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:17.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_8/index_7811_seed_14302_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:16:44.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 08:28:15.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_8/index_4784_seed_11872_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 08:29:40.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:40.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:30:55.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_8/index_4784_seed_11873_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:31:23.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 08:42:53.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_8/index_9116_seed_31515_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 08:44:19.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:19.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:45:35.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_8/index_9116_seed_31516_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:46:01.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:24.768 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 08:59:29.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_8/index_4133_seed_18133_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 09:00:53.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:53.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:10.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_8/index_4133_seed_18134_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:02:33.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 09:14:04.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_8/index_2951_seed_27919_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 09:15:28.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:28.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:16:52.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_8/index_2951_seed_27920_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:17:16.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 09:28:42.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_8/index_8822_seed_24840_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 09:30:06.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:30:06.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:27.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_8/index_8822_seed_24841_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:31:54.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 09:43:17.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_8/index_9546_seed_31698_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 09:44:39.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:39.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:02.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_8/index_9546_seed_31699_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:46:24.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 09:57:46.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:54.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_8/index_9393_seed_25840_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 09:59:13.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:59:13.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:23.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_8/index_9393_seed_25841_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:00:52.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 10:12:07.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_8/index_4579_seed_19364_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 10:13:31.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:31.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:14:44.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_8/index_4579_seed_19365_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 10:15:05.855 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:15:06.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 10:26:24.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_8/index_9914_seed_31989_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 10:27:44.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:44.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:02.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_8/index_9914_seed_31990_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:29:29.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 10:40:46.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:53.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_8/index_409_seed_32479_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 10:42:12.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:42:12.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:27.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_8/index_409_seed_32480_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:43:55.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 10:55:08.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:16.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_8/index_6941_seed_27111_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 10:56:35.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:36.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:57:52.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_8/index_6941_seed_27112_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 10:58:22.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:58:23.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 11:09:40.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:50.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_8/index_3340_seed_13175_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 11:11:07.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:11:07.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:20.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_8/index_3340_seed_13176_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:12:43.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:03.281 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 11:25:07.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_8/index_7413_seed_27927_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 11:26:36.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:36.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:27:52.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_8/index_7413_seed_27928_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 11:28:20.076 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:28:21.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 11:39:42.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_8/index_5919_seed_25503_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 11:41:10.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:41:10.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:19.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_8/index_5919_seed_25504_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 11:42:41.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:42:42.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 11:54:06.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:14.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_8/index_7749_seed_7514_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 11:55:29.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:29.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:56:42.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_8/index_7749_seed_7515_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:57:06.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 12:08:30.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_8/index_4689_seed_17535_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 12:09:50.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:50.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:07.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_8/index_4689_seed_17536_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 12:11:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:11:30.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 12:22:53.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_8/index_4328_seed_4482_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 12:24:17.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:17.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:30.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_8/index_4328_seed_4483_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:25:55.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 12:37:12.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_8/index_2556_seed_11510_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 12:38:36.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:36.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:39:47.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_8/index_2556_seed_11511_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 12:40:10.052 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:40:11.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 12:51:30.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_8/index_9877_seed_31823_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 12:52:56.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:56.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:10.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_8/index_9877_seed_31824_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:54:32.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 13:05:59.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_8/index_2970_seed_17713_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 13:07:20.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:20.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:08:38.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_8/index_2970_seed_17714_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:09:04.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 13:20:24.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:32.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_8/index_6433_seed_4096_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 13:21:47.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:47.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:00.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_8/index_6433_seed_4097_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:23:30.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 13:34:48.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_8/index_5255_seed_12308_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 13:36:13.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:36:13.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:25.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_8/index_5255_seed_12309_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:37:48.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.761 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 13:50:02.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_8/index_7373_seed_11020_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 13:51:30.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:30.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:52:48.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_8/index_7373_seed_11021_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:53:11.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 14:04:33.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_8/index_6659_seed_31298_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 14:05:59.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:59.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:19.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_8/index_6659_seed_31299_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:07:59.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 14:19:43.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_8/index_433_seed_18012_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 14:21:08.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:21:08.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:22:38.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_8/index_433_seed_18013_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:23:13.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 14:34:38.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_8/index_6894_seed_19191_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 14:36:00.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:36:00.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:16.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_8/index_6894_seed_19192_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:37:54.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 14:49:20.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:28.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_8/index_3724_seed_29833_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 14:50:42.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:42.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_8/index_3724_seed_29834_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:52:26.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 15:05:01.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:10.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_8/index_6960_seed_8185_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 15:06:25.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:25.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:07:40.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_8/index_6960_seed_8186_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:08:04.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 15:19:35.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_8/index_878_seed_9982_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 15:20:58.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:58.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:14.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_8/index_878_seed_9983_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:22:38.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 15:34:11.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_8/index_4540_seed_19685_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 15:35:43.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:43.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:36:53.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_8/index_4540_seed_19686_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:37:19.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 15:48:48.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:57.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_8/index_5029_seed_1741_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 15:50:15.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:50:15.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:25.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_8/index_5029_seed_1742_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:51:51.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 16:03:17.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_8/index_9842_seed_27896_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 16:04:41.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:41.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:05:57.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_8/index_9842_seed_27897_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:06:20.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.377 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 16:18:48.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_8/index_1827_seed_33124_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 16:20:15.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:15.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:21:33.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_8/index_1827_seed_33125_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:22:05.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 16:33:33.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:43.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_8/index_7361_seed_5679_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 16:34:59.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:59.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:16.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_8/index_7361_seed_5680_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:36:40.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 16:48:15.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:24.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_8/index_9119_seed_13055_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 16:49:43.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:43.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:50:53.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_8/index_9119_seed_13056_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:51:16.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 17:02:55.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:04.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_8/index_1906_seed_6123_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 17:04:19.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:19.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:33.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_8/index_1906_seed_6124_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:05:57.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 17:17:35.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_8/index_551_seed_16301_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 17:19:02.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:19:02.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:15.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_8/index_551_seed_16302_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:20:38.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 17:32:10.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_8/index_939_seed_8186_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 17:33:34.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:34.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:00.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_8/index_939_seed_8187_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:35:26.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 17:46:54.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_8/index_4544_seed_6927_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 17:48:19.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:19.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:49:38.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_8/index_4544_seed_6928_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:50:02.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 18:01:35.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:44.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_8/index_1786_seed_13719_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 18:03:00.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:03:00.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:17.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_8/index_1786_seed_13720_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:04:40.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 18:16:17.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_8/index_849_seed_29050_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 18:17:47.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:47.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:00.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_8/index_849_seed_29051_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:19:32.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.324 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 18:31:18.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:30.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_8/index_9333_seed_9617_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 18:32:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:45.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:14.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_8/index_9333_seed_9618_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 18:34:45.308 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:34:46.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:18.025 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 18:47:24.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_8/index_7181_seed_24127_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 18:48:49.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:49.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:12.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_8/index_7181_seed_24128_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 18:50:38.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:50:39.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 19:02:13.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_8/index_163_seed_16728_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 19:03:42.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:42.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:00.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_8/index_163_seed_16729_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:05:34.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 19:17:12.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:13.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:22.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_8/index_8266_seed_2309_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 19:18:39.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:39.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:19:58.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_8/index_8266_seed_2310_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:20:26.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 19:31:55.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:06.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_8/index_2303_seed_27464_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 19:33:20.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:20.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:34:55.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_8/index_2303_seed_27465_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 19:35:23.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:35:24.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 19:47:03.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_8/index_1307_seed_4463_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 19:48:29.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:29.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:49:47.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_8/index_1307_seed_4464_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:50:11.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 20:01:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:57.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_8/index_6733_seed_1214_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 20:03:14.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:03:14.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:29.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_8/index_6733_seed_1215_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:04:56.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 20:16:29.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:41.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_8/index_9815_seed_28846_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 20:17:55.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:55.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:19.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_8/index_9815_seed_28847_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:19:47.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 20:31:19.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_8/index_2714_seed_11864_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 20:32:45.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:45.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:00.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_8/index_2714_seed_11865_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:34:24.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 20:46:03.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_8/index_2926_seed_999_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 20:47:31.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:31.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:48:49.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_8/index_2926_seed_1000_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:49:15.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 21:00:47.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_8/index_8795_seed_1464_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 21:02:12.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:02:12.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:30.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_8/index_8795_seed_1465_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:03:58.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.475 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 21:17:31.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:41.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_8/index_2725_seed_9505_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 21:18:56.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:56.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:12.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_8/index_2725_seed_9506_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:20:38.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 21:32:09.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:51:09.712 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-30 21:51:09.712 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-30 21:51:10.406 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:51:10.406 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 21:55:22.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:49.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_8/index_1462_seed_13996_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 21:57:07.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:57:07.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:17.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_8/index_1462_seed_13997_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:59:08.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 34, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 22:10:52.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_8/index_5307_seed_18347_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 22:12:13.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:13.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:13:35.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_8/index_5307_seed_18348_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:14:03.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 22:25:33.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:41.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_8/index_1986_seed_27807_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 22:26:59.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:59.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:13.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_8/index_1986_seed_27808_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:28:37.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.656 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 22:40:03.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:04.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_8/index_159_seed_7234_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 22:41:27.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:27.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:42:45.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_8/index_159_seed_7235_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:43:10.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 22:54:38.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_8/index_4344_seed_4086_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 22:56:02.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:56:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:25.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_8/index_4344_seed_4087_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:57:50.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 23:09:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_8/index_9171_seed_24571_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 23:10:47.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:47.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:10.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_8/index_9171_seed_24572_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:12:36.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 23:24:05.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_8/index_3529_seed_31694_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 23:25:31.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:26:47.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_8/index_3529_seed_31695_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:27:14.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 23:38:46.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_8/index_1919_seed_23596_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 23:40:08.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:40:08.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:24.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_8/index_1919_seed_23597_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:41:48.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-30 23:53:10.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_8/index_7272_seed_13531_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-30 23:54:33.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:33.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:55:52.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_8/index_7272_seed_13532_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:56:15.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 00:07:41.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:52.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_8/index_9288_seed_24729_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 00:09:09.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:09:09.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:24.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_8/index_9288_seed_24730_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:10:46.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:53.960 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 00:22:59.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_8/index_1534_seed_7620_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 00:24:23.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:23.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:25:40.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_8/index_1534_seed_7621_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:26:05.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 00:37:30.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_8/index_6800_seed_18243_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 00:38:55.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:55.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:17.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_8/index_6800_seed_18244_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:40:46.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.326 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 00:52:07.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_8/index_2326_seed_13461_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 00:53:30.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:30.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:54:48.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_8/index_2326_seed_13462_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:55:22.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.395 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 01:06:44.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:52.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_8/index_5219_seed_26993_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 01:08:06.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:08:06.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_8/index_5219_seed_26994_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:09:52.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 01:21:16.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_8/index_7843_seed_21863_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 01:22:40.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:40.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:23:55.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_8/index_7843_seed_21864_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:24:20.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 01:35:36.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:43.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_8/index_296_seed_8939_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 01:37:00.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:37:00.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:15.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_8/index_296_seed_8940_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:38:38.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 01:50:03.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_8/index_7634_seed_15958_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 01:51:27.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:27.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:52:41.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_8/index_7634_seed_15959_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:53:06.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.331 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 02:04:34.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_8/index_6947_seed_26233_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 02:06:02.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:06:02.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:13.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_8/index_6947_seed_26234_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 02:07:38.764 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:07:39.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 34, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-05-31 02:19:01.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_8/index_3933_seed_7576_dp_17_rank_34_sample_2_slot_0.mp4 +2026-05-31 02:20:21.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:21.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:21:38.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_8/index_3933_seed_7577_dp_17_rank_34_sample_3_slot_1.mp4 +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:22:02.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:27.392 | INFO | __main__:__init__:946 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-06-25 18:01:27.395 | INFO | __main__:__init__:947 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-06-25 18:01:28.062 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:28.063 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 18:04:09.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_8/index_1462_seed_13996_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 18:05:50.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:50.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:06:56.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_8/index_1462_seed_13997_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:07:32.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 34, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 18:18:41.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_8/index_5307_seed_18347_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 18:20:00.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:20:00.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:12.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_8/index_5307_seed_18348_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:21:35.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 18:32:26.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_8/index_1986_seed_27807_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 18:33:48.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:48.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:34:54.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_8/index_1986_seed_27808_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:35:15.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 18:46:08.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_8/index_159_seed_7234_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 18:47:31.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:31.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:48:38.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_8/index_159_seed_7235_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 18:49:02.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:49:03.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 19:00:02.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_8/index_4344_seed_4086_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 19:01:24.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:24.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:34.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_8/index_4344_seed_4087_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:02:57.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 19:13:49.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:55.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_8/index_9171_seed_24571_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 19:15:15.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:15:15.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:20.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_8/index_9171_seed_24572_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:16:42.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 19:27:32.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_8/index_3529_seed_31694_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 19:28:52.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:52.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:03.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_8/index_3529_seed_31695_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:30:24.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 19:41:17.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_8/index_1919_seed_23596_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 19:42:36.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:36.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:43:50.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_8/index_1919_seed_23597_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:44:15.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 19:55:08.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_8/index_7272_seed_13531_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 19:56:29.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:29.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:57:41.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_8/index_7272_seed_13532_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:58:03.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 20:09:02.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_8/index_9288_seed_24729_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 20:10:25.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:25.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:35.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_8/index_9288_seed_24730_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 20:11:59.727 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:12:00.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:30.181 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 20:23:32.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_8/index_1534_seed_7620_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 20:24:59.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:59.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:05.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_8/index_1534_seed_7621_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:26:26.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 20:37:18.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_8/index_6800_seed_18243_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 20:38:38.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:38.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:39:48.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_8/index_6800_seed_18244_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:40:09.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 20:51:03.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_8/index_2326_seed_13461_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 20:52:24.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:24.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:35.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_8/index_2326_seed_13462_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:53:57.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 21:04:50.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:51.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:55.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_8/index_5219_seed_26993_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 21:06:09.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:06:09.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:23.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_8/index_5219_seed_26994_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 21:07:48.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:07:49.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 21:18:43.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_8/index_7843_seed_21863_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 21:20:06.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:20:06.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:17.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_8/index_7843_seed_21864_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:21:39.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 21:32:31.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_8/index_296_seed_8939_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 21:33:54.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:54.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:01.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_8/index_296_seed_8940_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 21:35:20.628 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:35:21.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 21:46:16.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_8/index_7634_seed_15958_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 21:47:37.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:37.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:48:47.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_8/index_7634_seed_15959_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:49:10.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 22:00:05.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_8/index_6947_seed_26233_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 22:01:24.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:24.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:36.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_8/index_6947_seed_26234_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:02:58.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 22:13:50.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_8/index_3933_seed_7576_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 22:15:09.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:15:09.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:22.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_8/index_3933_seed_7577_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:16:45.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 22:27:39.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:45.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_8/index_9019_seed_28616_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 22:29:03.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:29:03.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:10.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_8/index_9019_seed_28617_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:30:32.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:58.001 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 22:41:59.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:07.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_8/index_2759_seed_21861_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 22:43:21.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:21.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:31.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_8/index_2759_seed_21862_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:44:52.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.064 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 22:55:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_8/index_8546_seed_29910_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 22:57:07.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:57:07.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:16.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_8/index_8546_seed_29911_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:58:39.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 23:09:30.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_8/index_3707_seed_28676_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 23:10:49.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:49.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:01.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_8/index_3707_seed_28677_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:12:22.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 23:23:15.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_8/index_7427_seed_6728_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 23:24:38.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:38.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:25:47.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_8/index_7427_seed_6729_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:26:11.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 23:37:07.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_8/index_6551_seed_26444_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 23:38:29.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:29.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:39:37.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_8/index_6551_seed_26445_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:40:02.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-25 23:51:52.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_8/index_891_seed_33474_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-25 23:53:12.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:53:12.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:23.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_8/index_891_seed_33475_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:54:43.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 00:05:36.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:42.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_8/index_5807_seed_1832_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 00:06:59.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:59.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:07.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_8/index_5807_seed_1833_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:08:28.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 00:19:21.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_8/index_792_seed_22653_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 00:20:41.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:41.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:21:50.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_8/index_792_seed_22654_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:22:11.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 00:33:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:13.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_8/index_6045_seed_8793_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 00:34:31.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:31.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:35:37.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_8/index_6045_seed_8794_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:36:04.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 00:47:03.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:09.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_8/index_6794_seed_11833_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 00:48:24.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:24.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:37.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_8/index_6794_seed_11834_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:50:00.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:28.907 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 01:01:30.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:38.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_8/index_1880_seed_13841_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 01:02:55.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:55.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:03.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_8/index_1880_seed_13842_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:04:25.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 01:15:17.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_8/index_5534_seed_15000_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 01:16:39.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:39.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:17:50.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_8/index_5534_seed_15001_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:18:12.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 01:29:08.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:13.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_8/index_5095_seed_19288_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 01:30:30.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:31:40.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_8/index_5095_seed_19289_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:32:03.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 01:42:58.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_8/index_8455_seed_992_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 01:44:23.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:23.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:29.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_8/index_8455_seed_993_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:45:50.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 01:56:44.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_8/index_1388_seed_22577_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 01:58:10.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:58:10.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_8/index_1388_seed_22578_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:59:40.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 02:10:31.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_8/index_4030_seed_29169_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 02:11:55.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:55.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:03.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_8/index_4030_seed_29170_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:13:24.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 02:24:16.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_8/index_9229_seed_2480_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 02:25:34.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:34.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:26:46.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_8/index_9229_seed_2481_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:27:11.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 02:38:08.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_8/index_9708_seed_3187_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 02:39:30.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:30.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:36.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_8/index_9708_seed_3188_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 02:40:59.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:41:00.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 02:51:53.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_8/index_1904_seed_24828_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 02:53:17.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:53:17.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:25.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_8/index_1904_seed_24829_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:54:48.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 03:05:40.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:41.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_8/index_5748_seed_21478_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 03:07:10.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:07:10.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:15.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_8/index_5748_seed_21479_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:08:38.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.960 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 03:20:06.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_8/index_1943_seed_27500_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 03:21:27.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:27.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:39.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_8/index_1943_seed_27501_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:22:59.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 03:33:52.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_8/index_6584_seed_12338_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 03:35:15.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:35:15.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:23.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_8/index_6584_seed_12339_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:36:46.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 03:47:40.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_8/index_815_seed_18789_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 03:48:59.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:59.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:10.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_8/index_815_seed_18790_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:50:31.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 04:01:21.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:27.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_8/index_8934_seed_2551_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 04:02:40.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:40.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:03:51.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_8/index_8934_seed_2552_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:04:12.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:09.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 04:15:09.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_8/index_4320_seed_24277_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 04:16:30.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:30.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:17:42.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_8/index_4320_seed_24278_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:18:05.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 04:29:03.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:09.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_8/index_4988_seed_31733_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 04:30:24.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:24.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:33.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_8/index_4988_seed_31734_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:31:54.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 04:42:46.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:52.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_8/index_402_seed_29366_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 04:44:06.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:44:06.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:16.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_8/index_402_seed_29367_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:45:35.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.442 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 04:56:27.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_8/index_7811_seed_14301_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 04:57:53.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:53.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:58:58.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_8/index_7811_seed_14302_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:59:19.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.656 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 05:10:10.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_8/index_4784_seed_11872_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 05:11:31.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:31.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:12:41.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_8/index_4784_seed_11873_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:13:03.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 05:24:03.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_8/index_9116_seed_31515_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 05:25:26.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:26.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:35.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_8/index_9116_seed_31516_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:26:58.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.572 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.483 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 05:39:29.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_8/index_4133_seed_18133_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 05:40:50.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:50.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:00.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_8/index_4133_seed_18134_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:42:20.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 05:53:10.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:15.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_8/index_2951_seed_27919_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 05:54:33.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:33.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:39.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_8/index_2951_seed_27920_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:56:00.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 06:06:52.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_8/index_8822_seed_24840_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 06:08:14.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:08:14.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:22.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_8/index_8822_seed_24841_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:09:44.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 06:20:40.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:44.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_8/index_9546_seed_31698_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 06:21:58.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:58.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:09.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_8/index_9546_seed_31699_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 06:23:30.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:23:31.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 06:34:21.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_8/index_9393_seed_25840_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 06:35:46.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:46.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:36:51.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_8/index_9393_seed_25841_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 06:37:10.499 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:37:11.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 06:48:09.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:15.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_8/index_4579_seed_19364_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 06:49:29.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:29.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:50:44.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_8/index_4579_seed_19365_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:51:03.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 07:02:03.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_8/index_9914_seed_31989_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 07:03:21.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:21.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:33.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_8/index_9914_seed_31990_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:04:55.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 07:15:48.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:54.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_8/index_409_seed_32479_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 07:17:11.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:17:11.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:21.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_8/index_409_seed_32480_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:18:41.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 07:29:32.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_8/index_6941_seed_27111_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 07:30:58.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:58.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:04.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_8/index_6941_seed_27112_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:32:31.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 07:43:21.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:27.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_8/index_3340_seed_13175_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 07:44:42.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:42.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:45:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_8/index_3340_seed_13176_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:46:15.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.931 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 07:57:45.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:53.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_8/index_7413_seed_27927_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 07:59:12.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:59:12.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:18.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_8/index_7413_seed_27928_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:00:40.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 08:11:33.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_8/index_5919_seed_25503_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 08:12:52.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:52.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:02.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_8/index_5919_seed_25504_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:14:26.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 08:25:18.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_8/index_7749_seed_7514_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 08:26:42.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:42.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:27:47.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_8/index_7749_seed_7515_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:28:08.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 08:39:03.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_8/index_4689_seed_17535_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 08:40:24.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:24.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:35.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_8/index_4689_seed_17536_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:41:57.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 08:52:50.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_8/index_4328_seed_4482_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 08:54:11.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:54:11.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:22.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_8/index_4328_seed_4483_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:55:45.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 09:06:39.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_8/index_2556_seed_11510_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 09:08:01.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:08:01.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:11.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_8/index_2556_seed_11511_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:09:33.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 09:20:22.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:28.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_8/index_9877_seed_31823_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 09:21:44.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:44.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:22:52.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_8/index_9877_seed_31824_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:23:12.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 09:34:09.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:15.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_8/index_2970_seed_17713_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 09:35:30.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:30.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:36:41.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_8/index_2970_seed_17714_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:37:04.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 09:47:56.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_8/index_6433_seed_4096_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 09:49:18.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:18.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:29.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_8/index_6433_seed_4097_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 09:50:50.361 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:50:51.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 10:01:45.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_8/index_5255_seed_12308_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 10:03:09.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:03:09.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:15.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_8/index_5255_seed_12309_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:04:34.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.947 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 10:16:01.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_8/index_7373_seed_11020_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 10:17:27.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:27.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:34.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_8/index_7373_seed_11021_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:18:58.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 10:29:52.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_8/index_6659_seed_31298_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 10:31:11.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:31:11.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:23.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_8/index_6659_seed_31299_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:32:44.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 10:43:37.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_8/index_433_seed_18012_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 10:44:56.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:56.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:08.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_8/index_433_seed_18013_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:46:29.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 10:57:21.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_8/index_6894_seed_19191_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 10:58:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:41.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:59:56.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_8/index_6894_seed_19192_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:00:18.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 11:11:08.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_8/index_3724_seed_29833_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 11:12:29.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:29.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:39.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_8/index_3724_seed_29834_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:13:59.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.289 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 11:26:08.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_8/index_6960_seed_8185_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 11:27:29.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:29.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:28:40.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_8/index_6960_seed_8186_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:29:03.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 11:39:55.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_8/index_878_seed_9982_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 11:41:16.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:16.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:24.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_8/index_878_seed_9983_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:42:44.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 11:53:37.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_8/index_4540_seed_19685_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 11:54:57.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:57.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:07.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_8/index_4540_seed_19686_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:56:31.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 12:07:21.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:27.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_8/index_5029_seed_1741_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 12:08:41.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:41.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:09:51.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_8/index_5029_seed_1742_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:10:12.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 12:21:06.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_8/index_9842_seed_27896_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 12:22:26.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:26.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:38.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_8/index_9842_seed_27897_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:24:00.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:29.143 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 12:35:30.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:38.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_8/index_1827_seed_33124_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 12:36:55.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:55.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:03.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_8/index_1827_seed_33125_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:38:25.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 12:49:16.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_8/index_7361_seed_5679_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 12:50:41.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:41.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:51:48.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_8/index_7361_seed_5680_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:52:10.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 13:03:03.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_8/index_9119_seed_13055_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 13:04:26.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:26.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_8/index_9119_seed_13056_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:05:53.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 13:16:44.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_8/index_1906_seed_6123_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 13:18:08.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:18:08.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:16.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_8/index_1906_seed_6124_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:19:41.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 13:30:31.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_8/index_551_seed_16301_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 13:31:54.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:54.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:04.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_8/index_551_seed_16302_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:33:25.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 13:44:15.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_8/index_939_seed_8186_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 13:45:37.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:37.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:46:49.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_8/index_939_seed_8187_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:47:13.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 13:58:07.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:13.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_8/index_4544_seed_6927_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 13:59:29.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:29.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:36.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_8/index_4544_seed_6928_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:00:56.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 14:11:49.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:55.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_8/index_1786_seed_13719_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 14:13:12.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:13:12.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:18.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_8/index_1786_seed_13720_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:14:39.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 14:25:27.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_8/index_849_seed_29050_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 14:26:54.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:54.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:02.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_8/index_849_seed_29051_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:28:25.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.605 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 14:39:15.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_8/index_9333_seed_9617_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 14:40:37.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:37.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:41:50.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_8/index_9333_seed_9618_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:42:13.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.869 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 14:53:45.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:53.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_8/index_7181_seed_24127_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 14:55:09.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:55:09.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:16.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_8/index_7181_seed_24128_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:56:36.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 15:07:27.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_8/index_163_seed_16728_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 15:08:47.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:47.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:09:56.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_8/index_163_seed_16729_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:10:19.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 15:21:10.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:17.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_8/index_8266_seed_2309_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 15:22:34.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:34.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:23:41.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_8/index_8266_seed_2310_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 15:24:01.111 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:24:02.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 15:34:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_8/index_2303_seed_27464_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 15:36:15.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:15.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:29.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_8/index_2303_seed_27465_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:37:50.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 15:48:45.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_8/index_1307_seed_4463_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 15:50:05.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:50:05.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:15.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_8/index_1307_seed_4464_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:51:35.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 16:02:26.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_8/index_6733_seed_1214_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 16:03:48.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:48.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:04:57.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_8/index_6733_seed_1215_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:05:18.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 16:16:06.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:07.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_8/index_9815_seed_28846_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 16:17:25.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:25.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:18:42.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_8/index_9815_seed_28847_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:19:04.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 16:29:54.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_8/index_2714_seed_11864_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 16:31:15.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:15.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:23.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_8/index_2714_seed_11865_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:32:46.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 16:43:38.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_8/index_2926_seed_999_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 16:45:01.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:45:01.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:08.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_8/index_2926_seed_1000_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:46:31.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 16:57:20.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:21.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_8/index_8795_seed_1464_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 16:58:43.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:43.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:59:54.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_8/index_8795_seed_1465_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:00:15.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:50.844 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 17:12:53.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_8/index_2725_seed_9505_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 17:14:13.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:14:13.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:25.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_8/index_2725_seed_9506_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:15:48.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 17:26:39.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_8/index_4709_seed_14268_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 17:28:01.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:28:01.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:09.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_8/index_4709_seed_14269_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:29:30.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 17:40:20.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_8/index_6819_seed_11935_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 17:41:44.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:44.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:42:54.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_8/index_6819_seed_11936_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:43:14.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 17:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:12.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_8/index_4433_seed_3556_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 17:55:27.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:27.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:37.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_8/index_4433_seed_3557_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:56:58.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 18:07:49.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:55.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_8/index_9559_seed_1310_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 18:09:16.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:09:16.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:21.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_8/index_9559_seed_1311_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 18:10:44.980 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:10:46.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 18:21:38.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:44.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_8/index_8648_seed_11793_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 18:23:02.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:23:02.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:12.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_8/index_8648_seed_11794_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:24:37.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 18:35:27.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_8/index_4746_seed_261_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 18:36:48.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:48.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:03.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_8/index_4746_seed_262_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:38:24.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 18:49:15.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_8/index_5621_seed_21040_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 18:50:33.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:33.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:51:44.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_8/index_5621_seed_21041_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:52:05.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 19:03:02.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_8/index_4652_seed_29740_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 19:04:22.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:22.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:35.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_8/index_4652_seed_29741_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:05:55.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 19:16:46.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:52.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_8/index_1953_seed_18382_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 19:18:09.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:18:09.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:20.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_8/index_1953_seed_18383_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:19:40.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:30.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.680 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 19:31:09.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:17.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_8/index_1377_seed_29736_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 19:32:32.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:32.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:33:43.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_8/index_1377_seed_29737_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:34:07.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 19:45:03.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:09.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_8/index_8498_seed_7361_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 19:46:26.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:26.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:32.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_8/index_8498_seed_7362_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:47:52.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 19:58:45.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_8/index_9462_seed_28707_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 20:00:11.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:00:11.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:17.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_8/index_9462_seed_28708_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:01:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.572 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 20:12:28.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:34.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_8/index_1715_seed_33213_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 20:13:50.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:50.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:14:58.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_8/index_1715_seed_33214_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 20:15:21.128 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:15:22.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 20:26:14.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_8/index_5859_seed_23587_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 20:27:35.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:35.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:28:48.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_8/index_5859_seed_23588_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:29:12.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 20:40:07.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_8/index_3688_seed_3398_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 20:41:32.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:32.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:42:38.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_8/index_3688_seed_3399_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 20:43:00.483 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:43:01.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.373 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 20:53:52.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_8/index_7003_seed_27291_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 20:55:18.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:55:18.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:25.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_8/index_7003_seed_27292_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 20:56:53.176 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:56:54.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 21:07:44.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_8/index_9253_seed_31787_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 21:09:02.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:09:02.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:18.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_8/index_9253_seed_31788_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 21:10:41.549 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:10:42.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.484 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 21:21:31.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_8/index_4101_seed_22301_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 21:22:51.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:51.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:02.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_8/index_4101_seed_22302_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:24:26.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 21:35:16.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:16.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_8/index_7217_seed_24603_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 21:36:43.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:43.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:37:49.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_8/index_7217_seed_24604_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:38:11.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.557 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 21:49:43.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:51.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_8/index_8303_seed_9628_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 21:51:10.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:51:10.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:15.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_8/index_8303_seed_9629_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 21:52:37.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:52:38.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 22:03:26.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_8/index_1633_seed_27992_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 22:04:47.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:47.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:05:55.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_8/index_1633_seed_27993_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:06:17.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 22:17:07.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:13.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_8/index_2401_seed_30239_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 22:18:31.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:31.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:37.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_8/index_2401_seed_30240_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:19:59.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 22:30:49.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_8/index_5294_seed_32402_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 22:32:08.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:32:08.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:20.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_8/index_5294_seed_32403_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 22:33:42.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:33:43.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 22:44:32.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_8/index_9906_seed_15048_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 22:45:57.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:57.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:03.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_8/index_9906_seed_15049_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:47:24.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 22:59:16.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:22.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_8/index_7372_seed_1340_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 23:00:34.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:34.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:01:48.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_8/index_7372_seed_1341_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:02:09.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.731 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 23:13:02.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_8/index_2716_seed_12642_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 23:14:23.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:23.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:31.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_8/index_2716_seed_12643_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:15:53.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 23:26:44.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_8/index_3516_seed_21357_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 23:28:04.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:28:04.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:15.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_8/index_3516_seed_21358_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:29:39.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 23:40:27.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:33.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_8/index_5441_seed_18519_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 23:41:48.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:48.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:01.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_8/index_5441_seed_18520_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:43:26.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-26 23:54:16.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_8/index_2991_seed_25469_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-26 23:55:38.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:56:46.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_8/index_2991_seed_25470_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:57:09.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.246 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 00:08:40.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:48.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_8/index_2791_seed_463_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 00:10:05.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:10:05.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:11.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_8/index_2791_seed_464_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:11:32.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 00:22:20.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:26.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_8/index_6151_seed_20829_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 00:23:49.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:49.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:24:53.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_8/index_6151_seed_20830_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 00:25:14.490 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:25:15.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 00:36:06.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_8/index_8047_seed_17037_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 00:37:26.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:26.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:38.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_8/index_8047_seed_17038_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:38:59.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 00:49:49.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_8/index_8917_seed_30462_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 00:51:12.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:51:12.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:19.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_8/index_8917_seed_30463_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 00:52:43.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:52:44.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 01:03:35.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_8/index_2696_seed_33017_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 01:04:57.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:57.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:05.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_8/index_2696_seed_33018_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:06:28.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 01:17:19.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:20.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:24.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_8/index_8405_seed_692_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 01:18:38.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:38.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:19:52.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_8/index_8405_seed_693_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 01:20:14.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:20:15.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 01:31:07.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_8/index_3924_seed_10527_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 01:32:26.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:26.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:38.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_8/index_3924_seed_10528_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:33:59.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 01:44:47.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_8/index_4588_seed_29043_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 01:46:09.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:46:09.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:19.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_8/index_4588_seed_29044_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:47:40.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 01:58:27.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:33.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_8/index_5077_seed_14577_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 01:59:54.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:54.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:00.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_8/index_5077_seed_14578_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:01:23.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 02:12:14.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:15.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:21.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_8/index_6449_seed_21311_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 02:13:34.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:34.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:14:47.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_8/index_6449_seed_21312_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:15:09.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.552 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 02:26:39.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_8/index_5535_seed_15364_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 02:28:01.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:28:01.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:12.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_8/index_5535_seed_15365_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:29:34.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 02:40:23.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_8/index_640_seed_22598_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 02:41:45.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:45.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:00.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_8/index_640_seed_22599_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:43:25.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 02:54:14.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_8/index_2096_seed_2442_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 02:55:39.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:39.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:56:46.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_8/index_2096_seed_2443_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:57:09.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 03:08:01.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_8/index_9312_seed_6530_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 03:09:22.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:22.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:31.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_8/index_9312_seed_6531_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:10:50.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 03:21:41.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:47.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_8/index_3102_seed_9860_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 03:23:04.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:23:04.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:15.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_8/index_3102_seed_9861_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:24:38.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 03:35:26.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:32.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_8/index_3870_seed_30516_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 03:36:47.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:47.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:02.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_8/index_3870_seed_30517_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:38:25.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 03:49:14.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:20.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_8/index_5184_seed_13663_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 03:50:40.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:40.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:51:45.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_8/index_5184_seed_13664_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:52:08.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 04:03:03.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:09.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_8/index_8207_seed_5755_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 04:04:25.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:25.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:33.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_8/index_8207_seed_5756_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:05:54.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 04:16:48.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:54.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_8/index_1704_seed_3363_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 04:18:08.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:18:08.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:19.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_8/index_1704_seed_3364_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:19:42.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 04:30:30.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_8/index_175_seed_5065_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 04:31:52.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:52.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:01.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_8/index_175_seed_5066_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:33:22.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.259 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 04:45:58.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_8/index_6806_seed_17273_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 04:47:18.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:18.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:31.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_8/index_6806_seed_17274_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:48:56.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 04:59:46.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:53.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_8/index_4038_seed_11141_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 05:01:09.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:01:09.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:17.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_8/index_4038_seed_11142_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 05:02:39.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:02:40.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 05:13:27.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:33.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_8/index_3208_seed_19554_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 05:14:50.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:50.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:15:56.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_8/index_3208_seed_19555_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:16:17.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 05:27:07.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_8/index_5466_seed_33034_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 05:28:29.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:29.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:29:40.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_8/index_5466_seed_33035_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:30:01.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 05:40:50.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:56.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_8/index_4854_seed_30459_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 05:42:11.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:42:11.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:19.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_8/index_4854_seed_30460_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:43:44.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 05:54:35.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_8/index_6803_seed_30737_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 05:55:56.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:56.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:06.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_8/index_6803_seed_30738_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:57:30.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 06:08:19.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:25.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_8/index_8685_seed_27927_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 06:09:43.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:43.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:10:50.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_8/index_8685_seed_27928_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:11:14.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 06:22:07.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_8/index_4476_seed_9549_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 06:23:26.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:26.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:38.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_8/index_4476_seed_9550_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:24:59.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 06:35:50.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_8/index_5750_seed_15305_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 06:37:13.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:37:13.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:20.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_8/index_5750_seed_15306_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:38:42.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 06:49:31.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:36.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_8/index_2080_seed_10873_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 06:50:54.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:54.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:02.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_8/index_2080_seed_10874_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:52:21.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.833 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 07:03:49.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_8/index_7251_seed_21851_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 07:05:12.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:05:12.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:20.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_8/index_7251_seed_21852_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:06:40.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 07:17:29.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_8/index_6926_seed_9447_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 07:18:48.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:48.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:00.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_8/index_6926_seed_9448_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:20:22.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 07:31:16.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:22.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_8/index_7245_seed_13536_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 07:32:44.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:44.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:33:49.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_8/index_7245_seed_13537_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:34:12.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 07:45:07.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:13.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_8/index_3605_seed_19051_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 07:46:26.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:26.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:47:37.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_8/index_3605_seed_19052_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 07:48:01.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:48:02.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 07:58:52.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_8/index_6680_seed_25921_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 08:00:14.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:00:14.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:23.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_8/index_6680_seed_25922_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:01:46.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 08:12:38.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_8/index_2982_seed_18774_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 08:14:00.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:14:00.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:10.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_8/index_2982_seed_18775_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:15:31.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 08:26:20.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:26.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_8/index_3645_seed_9398_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 08:27:46.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:46.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:28:51.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_8/index_3645_seed_9399_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:29:11.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 08:40:03.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_8/index_1070_seed_2084_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 08:41:26.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:26.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:35.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_8/index_1070_seed_2085_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:42:57.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 08:53:47.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:54.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_8/index_2566_seed_30627_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 08:55:08.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:55:08.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:23.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_8/index_2566_seed_30628_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:56:50.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 09:07:41.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:47.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_8/index_2588_seed_12816_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 09:09:02.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:09:02.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:13.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_8/index_2588_seed_12817_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:10:37.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.019 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 09:22:02.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_8/index_3182_seed_6782_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 09:23:23.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:23.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:34.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_8/index_3182_seed_6783_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:24:55.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 09:35:43.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_8/index_3668_seed_5605_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 09:37:02.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:37:02.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:13.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_8/index_3668_seed_5606_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:38:35.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 09:49:26.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_8/index_1349_seed_2785_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 09:50:47.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:47.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:51:57.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_8/index_1349_seed_2786_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:52:18.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 10:03:08.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_8/index_1276_seed_17478_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 10:04:31.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:31.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:38.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_8/index_1276_seed_17479_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:06:00.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.252 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 10:16:53.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_8/index_3808_seed_17468_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 10:18:14.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:18:14.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:23.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_8/index_3808_seed_17469_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 10:19:45.587 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:19:46.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 10:31:49.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:49.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_8/index_663_seed_25959_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 10:33:10.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:33:10.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:21.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_8/index_663_seed_25960_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:34:43.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 10:45:31.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:32.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_8/index_3369_seed_26216_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 10:46:53.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:53.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:03.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_8/index_3369_seed_26217_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:48:27.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 10:59:16.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_8/index_1790_seed_32579_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 11:00:37.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:37.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:01:49.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_8/index_1790_seed_32580_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:02:13.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 11:13:07.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_8/index_5801_seed_23629_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 11:14:27.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:27.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:35.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_8/index_5801_seed_23630_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:15:55.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 11:26:47.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:53.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_8/index_8532_seed_28278_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 11:28:09.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:28:09.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:21.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_8/index_8532_seed_28279_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 11:29:40.798 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:29:42.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.615 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 11:41:06.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_8/index_2078_seed_6599_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 11:42:27.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:27.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:37.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_8/index_2078_seed_6600_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:43:59.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 11:54:48.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_8/index_7247_seed_14787_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 11:56:08.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:56:08.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:19.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_8/index_7247_seed_14788_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:57:45.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 12:08:34.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_8/index_1220_seed_2574_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 12:09:54.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:54.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:05.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_8/index_1220_seed_2575_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:11:27.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 12:22:19.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:25.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_8/index_1259_seed_32329_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 12:23:45.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:45.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:24:50.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_8/index_1259_seed_32330_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:25:10.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 12:36:01.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_8/index_7861_seed_3060_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 12:37:20.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:20.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:33.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_8/index_7861_seed_3061_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:38:53.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 12:49:44.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_8/index_2334_seed_22288_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 12:51:08.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:51:08.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:14.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_8/index_2334_seed_22289_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:52:34.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 13:03:21.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:27.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_8/index_3579_seed_16557_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 13:04:40.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:40.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:05:51.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_8/index_3579_seed_16558_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:06:13.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 13:17:07.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_8/index_951_seed_21464_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 13:18:30.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:30.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:37.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_8/index_951_seed_21465_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:19:58.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 13:30:47.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_8/index_9014_seed_9539_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 13:32:09.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:32:09.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:21.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_8/index_9014_seed_9540_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:33:47.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 13:44:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_8/index_5742_seed_21175_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 13:46:00.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:46:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:09.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_8/index_5742_seed_21176_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:47:31.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.283 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 13:58:57.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:03.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_8/index_3676_seed_28685_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 14:00:17.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:17.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:30.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_8/index_3676_seed_28686_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:01:51.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 14:12:43.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:49.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_8/index_8233_seed_10643_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 14:14:03.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:14:03.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:12.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_8/index_8233_seed_10644_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:15:32.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 14:26:20.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_8/index_158_seed_19480_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 14:27:44.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:44.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:28:50.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_8/index_158_seed_19481_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:29:12.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 14:40:05.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_8/index_9087_seed_6643_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 14:41:26.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:26.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:36.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_8/index_9087_seed_6644_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:42:58.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 14:53:47.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:53.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_8/index_4457_seed_1363_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 14:55:09.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:55:09.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:18.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_8/index_4457_seed_1364_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:56:39.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 15:07:26.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_8/index_3354_seed_22347_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 15:08:47.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:47.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:09:56.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_8/index_3354_seed_22348_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:10:18.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 15:21:06.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_8/index_5602_seed_18219_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 15:22:31.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:31.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:37.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_8/index_5602_seed_18220_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:23:58.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 15:34:47.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_8/index_1728_seed_12930_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 15:36:08.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:36:08.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:20.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_8/index_1728_seed_12931_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:37:44.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 15:48:32.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_8/index_1306_seed_4150_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 15:49:52.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:03.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_8/index_1306_seed_4151_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:51:25.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 16:02:14.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_8/index_9398_seed_770_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 16:03:32.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:32.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:04:44.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_8/index_9398_seed_771_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:05:05.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.677 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 16:17:43.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_8/index_2875_seed_22367_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 16:19:05.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:19:05.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:17.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_8/index_2875_seed_22368_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:20:37.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 16:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_8/index_5061_seed_28908_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 16:32:53.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:53.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:33:58.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_8/index_5061_seed_28909_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:34:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 16:45:07.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:13.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_8/index_6469_seed_24010_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 16:46:28.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:28.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:38.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_8/index_6469_seed_24011_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:47:59.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 16:58:47.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:53.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_8/index_4125_seed_657_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 17:00:08.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:00:08.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:20.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_8/index_4125_seed_658_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:01:42.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 17:12:31.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_8/index_4131_seed_17120_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 17:13:50.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:50.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:02.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_8/index_4131_seed_17121_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:15:22.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.262 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 17:26:15.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_8/index_369_seed_6862_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 17:27:34.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:34.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:28:46.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_8/index_369_seed_6863_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:29:09.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 17:40:00.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:06.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_8/index_6057_seed_28534_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 17:41:21.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:21.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:34.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_8/index_6057_seed_28535_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:42:55.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 17:53:45.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:51.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_8/index_7555_seed_9163_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 17:55:05.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:55:05.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:15.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_8/index_7555_seed_9164_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:56:38.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 18:07:25.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:31.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_8/index_6529_seed_6737_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 18:08:48.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:48.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:01.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_8/index_6529_seed_6738_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:10:24.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 18:21:14.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_8/index_4880_seed_13271_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 18:22:38.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:38.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:23:45.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_8/index_4880_seed_13272_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:24:07.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.642 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 18:35:38.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:44.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_8/index_3225_seed_10836_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 18:36:58.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:58.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:09.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_8/index_3225_seed_10837_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:38:32.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 18:49:21.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:27.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_8/index_7384_seed_1589_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 18:50:42.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:42.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:51:55.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_8/index_7384_seed_1590_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:52:16.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 19:03:06.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_8/index_5423_seed_30130_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 19:04:31.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:31.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:36.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_8/index_5423_seed_30131_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:05:55.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 19:16:46.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:52.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_8/index_5477_seed_1228_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 19:18:07.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:18:07.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:21.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_8/index_5477_seed_1229_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:19:42.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 19:30:30.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:36.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_8/index_3748_seed_17168_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 19:31:52.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:52.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:03.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_8/index_3748_seed_17169_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:33:25.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.246 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 19:44:14.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_8/index_7567_seed_746_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 19:45:33.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:33.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:46:44.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_8/index_7567_seed_747_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:47:10.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 19:58:02.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:08.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_8/index_9899_seed_29920_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 19:59:23.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:23.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:33.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_8/index_9899_seed_29921_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:00:54.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 20:11:43.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:43.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_8/index_4085_seed_23660_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 20:13:01.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:13:01.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:15.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_8/index_4085_seed_23661_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:14:39.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 20:25:26.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_8/index_5657_seed_16853_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 20:26:45.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:45.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:27:58.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_8/index_5657_seed_16854_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:28:19.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 20:39:06.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:07.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_8/index_2940_seed_31265_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 20:40:26.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:26.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:41:38.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_8/index_2940_seed_31266_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:42:04.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.439 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.043 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 20:53:29.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:35.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_8/index_2149_seed_21821_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 20:54:50.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:50.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:55:59.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_8/index_2149_seed_21822_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:56:21.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 21:07:10.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_8/index_1804_seed_6665_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 21:08:31.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:31.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:09:41.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_8/index_1804_seed_6666_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:10:07.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 21:21:01.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_8/index_2192_seed_1527_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 21:22:25.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:25.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:31.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_8/index_2192_seed_1528_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:23:52.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.975 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 21:34:42.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:48.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_8/index_2369_seed_25417_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 21:36:02.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:36:02.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:12.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_8/index_2369_seed_25418_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:37:34.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 21:48:23.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:23.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:30.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_8/index_5778_seed_19771_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 21:49:47.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:47.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:50:54.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_8/index_5778_seed_19772_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 21:51:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:51:15.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 22:03:21.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:27.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_8/index_7322_seed_4631_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 22:04:43.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:43.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:05:54.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_8/index_7322_seed_4632_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:06:17.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 22:17:06.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:12.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_8/index_4719_seed_25805_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 22:18:32.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:32.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:37.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_8/index_4719_seed_25806_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 22:19:56.152 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:19:57.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 22:30:47.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:53.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_8/index_7001_seed_21237_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 22:32:07.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:32:07.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:21.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_8/index_7001_seed_21238_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:33:41.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 22:44:30.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:36.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_8/index_6788_seed_13264_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 22:45:54.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:54.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:46:59.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_8/index_6788_seed_13265_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:47:18.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 22:58:07.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_8/index_8074_seed_28327_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 22:59:33.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:33.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:00:38.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_8/index_8074_seed_28328_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:01:01.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.671 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 23:12:29.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:35.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_8/index_3749_seed_20424_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 23:13:56.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:56.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:01.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_8/index_3749_seed_20425_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:15:23.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 23:26:15.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_8/index_4878_seed_29798_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 23:27:36.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:36.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:28:45.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_8/index_4878_seed_29799_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:29:06.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 23:40:02.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:08.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_8/index_8230_seed_16260_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 23:41:27.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:27.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:42:36.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_8/index_8230_seed_16261_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:43:01.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:50.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-27 23:53:51.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:57.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_8/index_3772_seed_15194_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-27 23:55:13.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:55:13.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:22.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_8/index_3772_seed_15195_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:56:44.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 00:07:32.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:38.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_8/index_8156_seed_19866_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 00:08:59.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:59.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:08.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_8/index_8156_seed_19867_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:10:31.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 00:21:21.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_8/index_2517_seed_14537_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 00:22:43.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:43.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:23:51.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_8/index_2517_seed_14538_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:24:12.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 00:35:07.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:13.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_8/index_9241_seed_10700_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 00:36:31.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:31.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:37:45.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_8/index_9241_seed_10701_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:38:10.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 00:49:02.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_8/index_311_seed_8664_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 00:50:24.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:24.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:36.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_8/index_311_seed_8665_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:51:57.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 01:02:48.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:54.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_8/index_1501_seed_7790_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 01:04:16.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:04:16.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:21.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_8/index_1501_seed_7791_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:05:44.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.806 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 01:16:33.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:39.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_8/index_5463_seed_24653_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 01:17:53.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:53.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:11.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_8/index_5463_seed_24654_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:19:31.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.434 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 01:30:58.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_8/index_5200_seed_31437_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 01:32:20.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:20.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:29.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_8/index_5200_seed_31438_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:33:52.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 01:44:44.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_8/index_1592_seed_16889_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 01:46:05.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:46:05.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_8/index_1592_seed_16890_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:47:37.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 01:58:27.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_8/index_81_seed_31680_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 01:59:47.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:47.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:00:58.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_8/index_81_seed_31681_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:01:18.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 02:12:08.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_8/index_8734_seed_10575_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 02:13:29.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:29.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:14:40.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_8/index_8734_seed_10576_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:15:03.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 02:25:53.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_8/index_1977_seed_12522_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 02:27:12.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:27:12.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:25.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_8/index_1977_seed_12523_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:28:45.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 02:39:38.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:44.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_8/index_752_seed_21355_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 02:40:58.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:58.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:08.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_8/index_752_seed_21356_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:42:31.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 02:53:20.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:21.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_8/index_4948_seed_5363_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 02:54:43.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:43.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:55:51.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_8/index_4948_seed_5364_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:56:13.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 03:07:08.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_8/index_6092_seed_15101_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 03:08:26.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:26.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:09:38.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_8/index_6092_seed_15102_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:10:03.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 03:20:53.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_8/index_4296_seed_13667_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 03:22:15.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:22:15.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:26.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_8/index_4296_seed_13668_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:23:49.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 03:34:40.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_8/index_7635_seed_13452_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 03:36:04.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:36:04.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:11.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_8/index_7635_seed_13453_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 03:37:34.480 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:37:35.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.356 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 03:50:13.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_8/index_4623_seed_9060_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 03:51:40.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:40.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:52:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_8/index_4623_seed_9061_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:53:06.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 04:04:03.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_8/index_117_seed_22947_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 04:05:25.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:25.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:33.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_8/index_117_seed_22948_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 04:06:55.563 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:06:56.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 04:17:48.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_8/index_9723_seed_29912_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 04:19:06.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:19:06.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:20.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_8/index_9723_seed_29913_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:20:47.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 04:31:40.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_8/index_6513_seed_6597_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 04:33:00.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:33:00.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_8/index_6513_seed_6598_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:34:29.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 04:45:20.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:26.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_8/index_492_seed_28662_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 04:46:40.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:40.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:47:50.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_8/index_492_seed_28663_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:48:11.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 04:59:03.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_8/index_9017_seed_26862_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 05:00:22.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:22.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:32.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_8/index_9017_seed_26863_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:01:52.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 05:12:44.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_8/index_3329_seed_15616_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 05:14:02.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:14:02.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:15.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_8/index_3329_seed_15617_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:15:35.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 05:26:25.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:32.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_8/index_2004_seed_1083_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 05:27:50.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:50.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:28:56.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_8/index_2004_seed_1084_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:29:22.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 05:40:15.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:21.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_8/index_1464_seed_20473_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 05:41:35.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:35.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:42:47.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_8/index_1464_seed_20474_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:43:10.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 05:54:06.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_8/index_9674_seed_21175_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 05:55:25.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:25.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:38.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_8/index_9674_seed_21176_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:57:00.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:25.125 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 06:08:27.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:33.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_8/index_4728_seed_4310_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 06:09:49.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:49.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:10:58.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_8/index_4728_seed_4311_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:11:21.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 06:22:11.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:16.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_8/index_6063_seed_30574_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 06:23:32.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:32.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:24:41.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_8/index_6063_seed_30575_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:25:07.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 06:36:02.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:08.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_8/index_139_seed_26859_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 06:37:23.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:23.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_8/index_139_seed_26860_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:38:53.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 06:49:45.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:51.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_8/index_5433_seed_3927_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 06:51:06.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:51:06.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:16.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_8/index_5433_seed_3928_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:52:37.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 07:03:27.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_8/index_376_seed_585_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 07:04:48.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:48.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:05:59.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_8/index_376_seed_586_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:06:21.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 07:17:12.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:18.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_8/index_6750_seed_31037_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 07:18:31.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:31.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:19:42.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_8/index_6750_seed_31038_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:20:04.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 07:30:53.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:59.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_8/index_1208_seed_18723_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 07:32:13.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:32:13.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:26.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_8/index_1208_seed_18724_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:33:48.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.442 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 07:44:40.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_8/index_9515_seed_5356_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 07:46:01.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:46:01.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:11.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_8/index_9515_seed_5357_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:47:33.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 07:58:21.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_8/index_1108_seed_10269_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 07:59:47.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:47.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:00:52.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_8/index_1108_seed_10270_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:01:16.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 08:12:08.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_8/index_4379_seed_17534_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 08:13:28.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:28.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:14:38.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_8/index_4379_seed_17535_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:15:03.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.488 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 08:26:31.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:37.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_8/index_2437_seed_11813_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 08:27:50.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:50.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:02.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_8/index_2437_seed_11814_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:29:27.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 08:40:22.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_8/index_3763_seed_20622_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 08:41:42.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:42.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:42:53.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_8/index_3763_seed_20623_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:43:14.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 08:54:07.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:14.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_8/index_9365_seed_19755_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 08:55:29.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:29.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:38.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_8/index_9365_seed_19756_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:57:00.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 09:07:50.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_8/index_4278_seed_1881_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 09:09:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:09:11.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:20.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_8/index_4278_seed_1882_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:10:44.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 09:21:32.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:38.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_8/index_9941_seed_26894_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 09:22:54.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:54.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:04.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_8/index_9941_seed_26895_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:24:28.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 09:36:35.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_8/index_7527_seed_5703_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 09:37:57.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:57.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:04.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_8/index_7527_seed_5704_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 09:39:26.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:39:27.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 09:50:22.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_8/index_8435_seed_12313_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 09:51:42.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:42.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:52:55.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_8/index_8435_seed_12314_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:53:25.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 10:04:16.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:22.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_8/index_7901_seed_27229_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 10:05:35.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:35.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:06:50.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_8/index_7901_seed_27230_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:07:11.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 10:18:09.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:15.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_8/index_8690_seed_13985_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 10:19:30.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:30.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:20:40.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_8/index_8690_seed_13986_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:21:02.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 10:31:54.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_8/index_6880_seed_16594_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 10:33:13.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:33:13.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:26.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_8/index_6880_seed_16595_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:34:54.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.683 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 10:46:20.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:27.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_8/index_4217_seed_20880_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 10:47:40.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:40.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:48:51.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_8/index_4217_seed_20881_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:49:13.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 11:00:08.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_8/index_865_seed_13442_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 11:01:26.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:26.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:39.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_8/index_865_seed_13443_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:03:00.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 11:13:51.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_8/index_7942_seed_12412_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 11:15:10.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:15:10.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:21.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_8/index_7942_seed_12413_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:16:43.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 11:27:32.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:38.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_8/index_5323_seed_7951_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 11:28:52.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:52.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:02.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_8/index_5323_seed_7952_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:30:23.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 11:41:15.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:21.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_8/index_3547_seed_24664_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 11:42:36.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:36.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:43:47.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_8/index_3547_seed_24665_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:44:08.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 11:55:03.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:09.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_8/index_5039_seed_2616_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 11:56:26.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:26.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:33.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_8/index_5039_seed_2617_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:57:57.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 12:08:48.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:54.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_8/index_5375_seed_11183_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 12:10:11.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:10:11.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:25.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_8/index_5375_seed_11184_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:11:48.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 12:22:40.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:46.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_8/index_1736_seed_24425_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 12:24:02.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:24:02.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:13.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_8/index_1736_seed_24426_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:25:34.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 12:36:24.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:30.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_8/index_4991_seed_32013_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 12:37:48.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:48.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:38:55.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_8/index_4991_seed_32014_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:39:17.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 12:50:08.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_8/index_6005_seed_25258_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 12:51:31.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:31.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:52:41.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_8/index_6005_seed_25259_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:53:04.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.808 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 13:04:33.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_8/index_826_seed_12527_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 13:05:58.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:58.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:07.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_8/index_826_seed_12528_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 13:07:33.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:07:33.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 13:18:22.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_8/index_6382_seed_21559_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 13:19:43.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:43.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:20:53.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_8/index_6382_seed_21560_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:21:14.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.326 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 13:32:07.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:13.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_8/index_1386_seed_30635_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 13:33:29.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:29.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:38.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_8/index_1386_seed_30636_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:35:00.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 13:45:49.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_8/index_500_seed_24372_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 13:47:09.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:47:09.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:22.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_8/index_500_seed_24373_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 13:48:41.544 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:48:42.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 13:59:30.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_8/index_2101_seed_27110_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 14:00:50.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:50.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:00.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_8/index_2101_seed_27111_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:02:21.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 14:13:08.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_8/index_3891_seed_12556_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 14:14:36.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:36.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:15:42.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_8/index_3891_seed_12557_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:16:09.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 14:27:02.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_8/index_5037_seed_3410_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 14:28:20.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:20.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:31.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_8/index_5037_seed_3411_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:29:51.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 14:40:43.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_8/index_8426_seed_1892_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 14:42:03.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:42:03.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:15.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_8/index_8426_seed_1893_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:43:35.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 14:54:25.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_8/index_1542_seed_32465_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 14:55:44.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:44.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:56:55.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_8/index_1542_seed_32466_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:57:15.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 15:08:07.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_8/index_9323_seed_9505_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 15:09:27.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:27.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:10:39.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_8/index_9323_seed_9506_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:11:02.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.164 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 15:23:36.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:42.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_8/index_5253_seed_5338_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 15:24:59.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:59.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:08.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_8/index_5253_seed_5339_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 15:26:31.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:26:32.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 15:37:21.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_8/index_1811_seed_14713_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 15:38:44.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:44.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:39:52.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_8/index_1811_seed_14714_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:40:15.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 15:51:08.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:14.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_8/index_5025_seed_31380_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 15:52:30.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:30.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:53:40.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_8/index_5025_seed_31381_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:54:05.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 16:04:54.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_8/index_7748_seed_20222_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 16:06:12.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:06:12.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:24.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_8/index_7748_seed_20223_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 16:07:49.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:07:50.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 16:18:40.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_8/index_3317_seed_24398_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 16:20:02.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:20:02.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:12.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_8/index_3317_seed_24399_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:21:34.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 16:32:23.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:29.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_8/index_5826_seed_22075_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 16:33:43.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:43.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:34:53.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_8/index_5826_seed_22076_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:35:14.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 16:46:07.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:13.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_8/index_12_seed_28736_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 16:47:28.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:28.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:39.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_8/index_12_seed_28737_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:49:00.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 16:59:50.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:56.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_8/index_4631_seed_21207_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 17:01:13.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:01:13.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:21.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_8/index_4631_seed_21208_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 17:02:43.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:02:44.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 17:13:34.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_8/index_1174_seed_1608_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 17:14:53.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:07.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_8/index_1174_seed_1609_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:16:32.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 17:27:21.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_8/index_6716_seed_29668_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 17:28:44.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:44.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:29:53.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_8/index_6716_seed_29669_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:30:18.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.859 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 17:41:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_8/index_2398_seed_1880_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 17:43:05.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:43:05.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:20.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_8/index_2398_seed_1881_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:44:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.909 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 17:55:31.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_8/index_6639_seed_3752_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 17:56:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:50.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:02.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_8/index_6639_seed_3753_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:58:25.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 18:09:15.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_8/index_6706_seed_4722_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 18:10:33.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:33.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:11:49.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_8/index_6706_seed_4723_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:12:10.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 18:23:03.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_8/index_6568_seed_229_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 18:24:22.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:22.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:33.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_8/index_6568_seed_230_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 18:25:53.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:25:54.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 18:36:44.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:44.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_8/index_6855_seed_19384_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 18:38:06.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:38:06.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:16.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_8/index_6855_seed_19385_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:39:39.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 18:50:26.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:32.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_8/index_4867_seed_27881_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 18:51:49.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:49.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:52:59.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_8/index_4867_seed_27882_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:53:22.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 19:04:13.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:14.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_8/index_4490_seed_27544_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 19:05:36.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:36.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:06:45.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_8/index_4490_seed_27545_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 19:07:07.109 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:07:08.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 19:18:02.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_8/index_3108_seed_24507_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 19:19:23.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:23.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:30.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_8/index_3108_seed_24508_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:20:53.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 19:31:44.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:50.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_8/index_7693_seed_2386_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 19:33:07.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:33:07.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:16.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_8/index_7693_seed_2387_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 19:34:36.662 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:34:37.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 19:45:24.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:29.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_8/index_2099_seed_14769_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 19:46:43.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:43.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:47:56.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_8/index_2099_seed_14770_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 19:48:17.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:48:18.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.743 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 19:59:44.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_8/index_2175_seed_29943_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 20:01:06.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:01:06.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:14.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_8/index_2175_seed_29944_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:02:35.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 20:13:25.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_8/index_5364_seed_16724_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 20:14:46.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:46.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:15:55.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_8/index_5364_seed_16725_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:16:16.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 20:27:07.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_8/index_7949_seed_6857_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 20:28:26.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:26.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:38.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_8/index_7949_seed_6858_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:29:59.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 20:40:49.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_8/index_9053_seed_2052_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 20:42:11.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:42:11.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:22.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_8/index_9053_seed_2053_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:43:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 20:54:36.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:42.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_8/index_4006_seed_28772_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 20:55:59.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:59.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:09.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_8/index_4006_seed_28773_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:57:34.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 21:09:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_8/index_1751_seed_20332_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 21:11:02.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:11:02.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:14.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_8/index_1751_seed_20333_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:12:35.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 21:23:24.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_8/index_5580_seed_5371_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 21:24:43.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:43.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:25:53.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_8/index_5580_seed_5372_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 21:26:14.845 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:26:15.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 21:37:07.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_8/index_9819_seed_8154_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 21:38:28.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:28.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:36.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_8/index_9819_seed_8155_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:39:58.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 21:50:48.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_8/index_4414_seed_641_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 21:52:08.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:52:08.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:21.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_8/index_4414_seed_642_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:53:43.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 22:04:31.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_8/index_3438_seed_28254_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 22:05:52.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:52.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:04.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_8/index_3438_seed_28255_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:07:26.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.850 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 22:18:53.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_8/index_1287_seed_12420_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 22:20:15.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:20:15.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:24.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_8/index_1287_seed_12421_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:21:47.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 22:32:40.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_8/index_1774_seed_30141_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 22:34:03.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:34:03.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:11.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_8/index_1774_seed_30142_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:35:35.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 22:46:24.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_8/index_1959_seed_27184_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 22:47:45.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:48:56.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_8/index_1959_seed_27185_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:49:18.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 23:00:07.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_8/index_520_seed_15884_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 23:01:29.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:29.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:38.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_8/index_520_seed_15885_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 23:02:58.268 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:02:59.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 23:13:49.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:50.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_8/index_445_seed_10798_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 23:15:09.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:15:09.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:20.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_8/index_445_seed_10799_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:16:44.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 23:27:32.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_8/index_4471_seed_31187_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 23:28:54.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:54.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:02.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_8/index_4471_seed_31188_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:30:23.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 23:41:14.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_8/index_1039_seed_27920_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 23:42:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:32.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:43:44.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_8/index_1039_seed_27921_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 23:44:05.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:44:06.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-28 23:55:00.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_8/index_6079_seed_25485_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-28 23:56:18.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:19.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:28.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_8/index_6079_seed_25486_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:57:49.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-29 00:08:39.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:45.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_8/index_3234_seed_20850_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-29 00:10:02.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:10:02.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:08.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_8/index_3234_seed_20851_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:11:29.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-29 00:22:19.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:25.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_8/index_4469_seed_24687_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-29 00:23:40.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:40.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:24:52.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_8/index_4469_seed_24688_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:25:13.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.283 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.085 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-29 00:36:43.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:49.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_8/index_8069_seed_9159_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-29 00:38:06.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:38:06.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:15.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_8/index_8069_seed_9160_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-29 00:39:38.559 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:39:39.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-29 00:50:27.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:33.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_8/index_8769_seed_18374_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-29 00:51:48.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:48.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:52:56.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_8/index_8769_seed_18375_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-29 00:53:18.756 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:53:19.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 34, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 34, dp_rank 17: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=8, rank_in_group=1 +2026-06-29 01:04:09.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_8/index_9856_seed_10846_dp_17_rank_34_sample_2_slot_0.mp4 +2026-06-29 01:05:32.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 34 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:32.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 34 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:06:40.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 34: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_8/index_9856_seed_10847_dp_17_rank_34_sample_3_slot_1.mp4 +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 34: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 01:07:03.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 34: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 34: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 34, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_41.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_41.log new file mode 100644 index 0000000000000000000000000000000000000000..a412136bbce76f2318c232c59558402eca42dd0b --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_41.log @@ -0,0 +1,5147 @@ +2026-05-29 20:46:54.962 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:46:54.966 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 20:51:46.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:12.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 41, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 21:06:57.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:02.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 21:21:17.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:15.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 21:35:35.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:42.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:42.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 21:49:56.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:04.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.289 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 22:04:18.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 22:18:35.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:41.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 22:32:58.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:01.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 22:47:20.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:28.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 23:01:44.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:52.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:49.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:54.965 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 23:16:57.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:57.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 23:31:14.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 23:45:29.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:35.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:35.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-29 23:59:50.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:57.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 00:14:13.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 00:28:36.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:45.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:57.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 00:42:58.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:06.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 00:57:15.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:21.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:21.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 01:11:38.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 01:26:05.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:11.661 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 01:41:19.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:28.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:18.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 01:55:38.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:44.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:39.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 02:09:58.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:03.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 02:24:23.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:31.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 02:38:56.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:03.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:58.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 02:54:31.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:38.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 03:08:50.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:57.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:53.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 03:23:09.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:16.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:16.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 03:37:26.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 03:51:59.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:06.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:16.675 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 04:07:19.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 04:21:48.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:58.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 04:36:12.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 04:50:41.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 05:05:07.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:13.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:12.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 05:19:25.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 05:33:38.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:44.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:44.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:37.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 05:47:59.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:58.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 06:02:21.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 06:16:55.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 06:20:00.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:13.135 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 06:32:17.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 06:46:37.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:43.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:38.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 07:00:59.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:07.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:00.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.615 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 07:15:21.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:22.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:42.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 07:30:08.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 07:44:30.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:38.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:59.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 07:59:00.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:08.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:33.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 08:13:33.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:41.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 08:28:15.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:24.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:24.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 08:42:53.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:02.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:26.833 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 08:59:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:39.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:33.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 09:14:05.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:12.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:12.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:17.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 09:28:42.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:51.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 09:43:17.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:25.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 09:57:46.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:53.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 10:12:07.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:06.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 10:26:24.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:31.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:45.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 10:40:46.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:54.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 10:55:08.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:09.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:17.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 11:09:40.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:48.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:48.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:08.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:05.189 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 11:25:08.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:17.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:21.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 11:39:42.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:50.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:05.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 11:54:06.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 12:08:30.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:37.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:30.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 12:22:53.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:01.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:01.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 12:37:12.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:11.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 12:51:30.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.549 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:58.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 13:05:59.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:06.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:05.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 13:20:24.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:32.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:32.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:30.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 13:34:48.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.740 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 13:50:02.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:10.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:10.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 14:04:33.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:34.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 14:19:43.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:55.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 14:34:38.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 14:49:20.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:27.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 15:05:01.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:10.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:10.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:04.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:34.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 15:19:35.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 15:34:11.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:21.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.806 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 15:48:48.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:51.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 16:03:17.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:25.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:46.325 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 16:18:48.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:01.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 16:33:33.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:33.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 16:48:15.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 17:02:55.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 17:17:35.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:44.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 17:32:10.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 17:46:54.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:03.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 18:01:35.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 18:16:18.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 18:31:18.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:19.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:20.896 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 18:47:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:40.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.245 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 19:02:13.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:35.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 19:17:12.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:27.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 19:31:55.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:02.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 19:47:03.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:12.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:45.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 20:01:46.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 20:16:29.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 20:31:19.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:02.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 20:46:03.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:12.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:47.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 21:00:47.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:28.332 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 21:17:31.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:41.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:41.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:38.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 21:32:09.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:54:34.406 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:54:34.407 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 21:55:22.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:48.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:48.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:08.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 41, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 22:10:52.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:53.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:11:59.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 22:25:33.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:42.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:42.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 22:40:03.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:12.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:11.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 22:54:39.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:46.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:50.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:19.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 23:09:20.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:28.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 23:24:05.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:15.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 23:38:46.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:54.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:09.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-30 23:53:10.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:18.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:16.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 00:07:41.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:53.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:05.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:57.147 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 00:22:59.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 00:37:30.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:39.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 00:52:07.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:14.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:22.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 01:06:44.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:53.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 01:21:16.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 01:35:36.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:44.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 01:50:03.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:11.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:11.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:07.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:33.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 02:04:34.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:42.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:40.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 41, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-05-31 02:19:01.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:08.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:03.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:05.447 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:05.447 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 18:04:09.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:31.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 41, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 18:18:41.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:35.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.711 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 18:32:26.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 18:46:08.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:13.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:03.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 19:00:03.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 19:13:49.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 19:27:32.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:24.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 19:41:16.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:22.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 19:55:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:13.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:13.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.17s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 20:09:02.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.742 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 20:23:32.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 20:37:18.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:19.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:24.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 20:51:03.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:57.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 21:04:50.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:55.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:55.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 21:18:43.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:48.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 21:32:31.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 21:46:16.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 22:00:05.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:11.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 22:13:50.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:55.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:55.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 22:27:39.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:44.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:58.190 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 22:41:59.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.063 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 22:55:45.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:51.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:51.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 23:09:30.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:35.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 23:23:15.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:20.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:11.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 23:37:07.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:12.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:02.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-25 23:51:52.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 00:05:36.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 00:19:21.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:25.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:25.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 00:33:07.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 00:47:03.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:00.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:29.039 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 01:01:30.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 01:15:17.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:17.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:23.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:12.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 01:29:08.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:13.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:13.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 01:42:58.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 01:56:44.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 02:10:31.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 02:24:16.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 02:38:08.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:13.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 02:41:00.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 02:51:53.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:54.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:59.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 03:05:40.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:45.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:45.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.781 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 03:20:06.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 03:33:52.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:57.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:57.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 03:47:40.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:45.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 04:01:21.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 04:15:09.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 04:29:03.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 04:42:46.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 04:56:27.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:32.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:32.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 05:10:10.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:03.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 05:24:03.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.014 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 05:39:29.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:35.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 05:53:10.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:15.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:15.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 06:06:52.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 06:20:39.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:45.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 06:34:21.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:27.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 06:48:09.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:14.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 07:02:03.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:55.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 07:15:48.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 07:29:32.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:37.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:30.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 07:43:21.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.865 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 07:57:46.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:52.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 08:11:33.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 08:25:18.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 08:39:03.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:57.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 08:52:50.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:55.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 09:06:39.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 09:20:22.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:28.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:28.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 09:34:09.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:14.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 09:47:56.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:02.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:51.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 10:01:45.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:34.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:16:00.013 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 10:16:01.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:59.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 10:29:52.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:44.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 10:43:37.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:29.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 10:57:21.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 11:11:08.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:09.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:13.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:13.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.283 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 11:26:08.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:13.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.143 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.252 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 11:39:55.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 11:53:37.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:42.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:31.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 12:07:21.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:12.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 12:21:06.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:07.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:12.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.907 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 12:35:30.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:37.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:37.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 12:49:16.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:21.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.745 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:10.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 13:03:03.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:07.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:07.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:53.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 13:16:44.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:49.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 13:30:31.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 13:44:15.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 13:58:07.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 14:11:49.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:39.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 14:25:28.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:32.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:32.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:25.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 14:39:15.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:21.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:13.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.246 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 14:53:45.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 15:07:27.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 15:21:11.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:16.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 15:34:56.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 15:48:45.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:50.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 16:02:26.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:31.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:31.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:17.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:05.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 16:16:06.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:07.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:12.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:04.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 16:29:54.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:00.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 16:43:38.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 16:57:20.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:26.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.143 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 17:12:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:47.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 17:26:39.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:44.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 17:40:20.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 17:54:06.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 18:07:49.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 18:21:38.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:37.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 18:35:27.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:33.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 18:49:15.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:20.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:05.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 19:03:02.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:07.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 19:16:46.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:51.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.593 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 19:31:08.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:07.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.122 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 19:45:03.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 19:58:45.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:51.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:37.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 20:12:28.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:22.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 20:26:14.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 20:40:07.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:12.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 20:53:52.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:58.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:54.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 21:07:44.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 21:21:31.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 21:35:16.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:22.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:11.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.477 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 21:49:43.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:43.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 22:03:26.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:27.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:32.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:18.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 22:17:07.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 22:30:49.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 22:44:32.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 22:59:16.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 23:13:02.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:07.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:07.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:53.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 23:26:44.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 23:40:27.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:26.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-26 23:54:16.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.402 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 00:08:40.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 00:22:20.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 00:36:06.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:59.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 00:49:49.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 01:03:35.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:40.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 01:17:19.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:20.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:25.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 01:31:06.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:06.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:12.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:58.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 01:44:47.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:52.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:52.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 01:58:27.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 02:12:14.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:36.913 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 02:26:39.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:33.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 02:40:23.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:28.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 02:54:14.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:09.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 03:08:01.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:07.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:50.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 03:21:41.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:46.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:46.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 03:35:26.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 03:49:14.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 04:03:03.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:08.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 04:16:48.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:53.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 04:30:31.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:35.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:35.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.537 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 04:45:58.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:05.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 04:59:46.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:51.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:51.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:39.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 05:13:28.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:33.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:33.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 05:27:07.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:13.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 05:40:50.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 05:54:35.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:40.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 06:08:19.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:14.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 06:22:07.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:13.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 06:35:50.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:55.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 06:49:30.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.829 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 07:03:49.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:56.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 07:17:29.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:34.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:23.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.395 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 07:31:16.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:12.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 07:45:07.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 07:58:52.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:57.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 08:12:38.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:43.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 08:26:20.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:10.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 08:40:03.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:08.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 08:53:47.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:53.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 09:07:41.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:46.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.171 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 09:22:03.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 09:35:43.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:48.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 09:49:26.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:31.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:17.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 10:03:08.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:13.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.863 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 10:16:53.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:58.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 10:31:49.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:54.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:54.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 10:45:31.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 10:59:16.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:21.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:21.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:12.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 11:13:07.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:12.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.423 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 11:26:47.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:41.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.060 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 11:41:06.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:12.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:59.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 11:54:48.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:52.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:52.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 12:08:34.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:39.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 12:22:19.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 12:36:01.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 12:49:44.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 13:03:21.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 13:17:07.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:13.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.373 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 13:30:47.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:47.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 13:44:38.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:43.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.220 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.189 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 13:58:57.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:03.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:03.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 14:12:43.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 14:26:20.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:25.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:12.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 14:40:05.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:06.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:11.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:58.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 14:53:47.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 15:07:26.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:32.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 15:21:06.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:11.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:58.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 15:34:47.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:47.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:53.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 15:48:32.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:37.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 16:02:14.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:05.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.661 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 16:17:43.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:50.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 16:31:26.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 16:45:07.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:59.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 16:58:47.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:52.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:41.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 17:12:31.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 17:26:15.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:20.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:09.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 17:40:00.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:00.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:05.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 17:53:45.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:38.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 18:07:25.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:24.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 18:21:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:20.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:07.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.367 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 18:35:38.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:43.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 18:49:21.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 19:03:06.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:11.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 19:16:46.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:51.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 19:30:30.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:25.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 19:44:14.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:19.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:10.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 19:58:02.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:53.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 20:11:43.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:49.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:39.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 20:25:26.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:19.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 20:39:06.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:07.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:12.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.411 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 20:53:29.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:34.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.307 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:21.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 21:07:10.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:15.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:07.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 21:21:01.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:51.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 21:34:42.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 21:48:23.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:23.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 22:03:21.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:16.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 22:17:06.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:07.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:57.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 22:30:47.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:52.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 22:44:30.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 22:58:07.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:12.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.779 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 23:12:29.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:36.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 23:26:15.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:20.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 23:40:02.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.289 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-27 23:53:51.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:56.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 00:07:32.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 00:21:21.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:27.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 00:35:07.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 00:49:02.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:07.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:57.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 01:02:48.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 01:16:33.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.951 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 01:30:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 01:44:44.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 01:58:27.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 02:12:08.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:13.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:03.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 02:25:53.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 02:39:38.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:31.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 02:53:20.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:26.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 03:07:08.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:03.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 03:20:53.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:48.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 03:34:40.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:45.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.713 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 03:50:13.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 04:04:03.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 04:17:48.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:54.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.524 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 04:31:40.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 04:45:20.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:10.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 04:59:03.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 05:12:44.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:45.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:50.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:34.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 05:26:26.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:22.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 05:40:15.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 05:54:06.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:07.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:12.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.856 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 06:08:27.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:21.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 06:22:11.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:17.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:07.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 06:36:02.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:07.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 06:49:45.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 07:03:27.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 07:17:12.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:17.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:04.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 07:30:53.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:48.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 07:44:40.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:32.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 07:58:21.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:27.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 08:12:08.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:13.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.750 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 08:26:31.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 08:40:22.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:28.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:14.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 08:54:07.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:13.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 08:57:00.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 09:07:50.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:56.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 09:21:32.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:37.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 09:36:35.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 09:50:22.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:25.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 10:04:16.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 10:18:09.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:14.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:02.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 10:31:54.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:54.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.741 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 10:46:20.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 11:00:08.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 11:13:51.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 11:27:32.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 11:41:15.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 11:55:03.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 12:08:48.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:53.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:48.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 12:22:40.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 12:36:24.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 12:50:08.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:13.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:13.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:31.120 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 13:04:33.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:34.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 13:18:22.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:28.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:14.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 13:32:07.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 13:45:49.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:49.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:55.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 13:59:30.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:36.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:21.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 14:13:08.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 14:27:02.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 14:40:44.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:49.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 14:54:25.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 15:08:07.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:12.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:02.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.535 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 15:23:36.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:32.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 15:37:21.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 15:51:08.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 16:04:54.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:50.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 16:18:40.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:41.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:45.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:45.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 16:32:23.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:28.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 16:46:07.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 16:59:50.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 17:13:34.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 17:27:21.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:25.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:25.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.673 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 17:41:44.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:49.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:49.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 17:55:31.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 18:09:15.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:21.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 18:23:03.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 18:36:44.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:44.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:50.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:39.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 18:50:26.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:22.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 19:04:13.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:14.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:19.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:08.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 19:18:02.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:07.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 19:31:44.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 19:45:24.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:24.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:30.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.295 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 19:59:44.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 20:13:24.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:24.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:30.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:16.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 20:27:07.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:12.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 20:40:49.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:54.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 20:54:36.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 21:09:41.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:47.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 21:23:23.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:29.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 21:37:07.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 21:50:48.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:53.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:53.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 22:04:31.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:31.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:37.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.943 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 22:18:53.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 22:32:40.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:35.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 22:46:24.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:29.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 23:00:07.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:59.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 23:13:49.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:55.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:55.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 23:27:32.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:23.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 23:41:14.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-28 23:55:00.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.655 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-29 00:08:39.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:29.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-29 00:22:19.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.527 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-29 00:36:43.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:48.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-29 00:50:27.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 41, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 41, dp_rank 20: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=10, rank_in_group=0 +2026-06-29 01:04:09.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 41 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:14.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 41 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 41: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 41: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 41: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 41, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_46.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_46.log new file mode 100644 index 0000000000000000000000000000000000000000..a0a415667755429fdbef7e2fba405b0253023d76 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_46.log @@ -0,0 +1,6079 @@ +2026-05-29 20:47:18.346 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-29 20:47:18.346 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-29 20:47:19.082 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:47:19.082 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 20:51:46.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_11/index_1871_seed_32875_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 20:53:32.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:32.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:38.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_11/index_1871_seed_32876_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 20:55:17.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 46, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 21:06:57.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:04.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_11/index_3841_seed_712_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 21:08:21.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:21.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:09:39.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_11/index_3841_seed_713_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:10:02.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 21:21:17.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:25.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_11/index_3109_seed_10553_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 21:22:42.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:42.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:23:53.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_11/index_3109_seed_10554_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:24:16.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 21:35:35.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_11/index_6414_seed_17502_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 21:36:58.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:58.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:14.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_11/index_6414_seed_17503_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:38:36.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 21:49:56.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_11/index_5259_seed_2698_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 21:51:23.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:23.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:52:35.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_11/index_5259_seed_2699_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:53:01.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 22:04:18.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:26.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_11/index_6402_seed_15149_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 22:05:43.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:43.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:06:52.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_11/index_6402_seed_15150_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:07:15.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 22:18:35.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:42.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_11/index_5821_seed_30564_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 22:19:57.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:57.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:14.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_11/index_5821_seed_30565_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:21:38.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 22:32:59.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:06.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_11/index_6055_seed_9340_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 22:34:20.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:20.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:36.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_11/index_6055_seed_9341_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:36:01.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 22:47:20.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_11/index_8903_seed_12168_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 22:48:49.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:49.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:49:56.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_11/index_8903_seed_12169_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:50:20.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 23:01:44.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_11/index_8562_seed_20162_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 23:03:06.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:03:06.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:23.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_11/index_8562_seed_20163_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:04:49.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.550 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:54.672 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 23:16:57.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_11/index_3672_seed_32787_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 23:18:24.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:24.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:35.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_11/index_3672_seed_32788_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 23:19:57.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:19:58.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:13.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 23:31:14.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:21.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_11/index_2179_seed_9286_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 23:32:37.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:37.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:33:52.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_11/index_2179_seed_9287_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:34:15.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 23:45:29.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_11/index_4971_seed_26688_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-29 23:46:50.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:50.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:09.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_11/index_4971_seed_26689_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:48:31.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-29 23:59:50.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:58.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_11/index_8020_seed_17272_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 00:01:15.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:01:15.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:26.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_11/index_8020_seed_17273_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 00:02:54.223 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:02:55.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 00:14:13.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:20.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_11/index_7420_seed_33081_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 00:15:39.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:39.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:16:51.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_11/index_7420_seed_33082_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:17:18.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 00:28:36.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_11/index_2730_seed_16331_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 00:30:04.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:30:04.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_11/index_2730_seed_16332_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 00:31:38.351 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:31:39.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 00:42:58.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_11/index_4503_seed_14774_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 00:44:21.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:21.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:45:38.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_11/index_4503_seed_14775_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:46:02.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 00:57:15.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_11/index_3263_seed_17977_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 00:58:35.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:35.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:59:51.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_11/index_3263_seed_17978_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:00:16.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 01:11:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:45.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_11/index_9452_seed_7775_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 01:13:05.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:13:05.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:18.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_11/index_9452_seed_7776_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:14:44.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.811 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 01:26:05.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:14.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_11/index_1765_seed_27681_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 01:27:31.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:31.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:28:42.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_11/index_1765_seed_27682_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 01:29:03.428 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:29:04.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:10.651 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 01:41:19.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_11/index_7039_seed_23291_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 01:42:48.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:48.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:43:56.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_11/index_7039_seed_23292_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:44:18.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 01:55:38.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:45.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_11/index_856_seed_10285_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 01:57:05.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:57:05.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:14.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_11/index_856_seed_10286_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:58:40.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 02:09:58.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_11/index_9375_seed_29287_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 02:11:21.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:21.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:12:39.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_11/index_9375_seed_29288_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:13:03.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 02:24:23.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_11/index_5032_seed_22409_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 02:25:54.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:54.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:05.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_11/index_5032_seed_22410_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:27:33.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 02:38:56.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:02.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_11/index_5849_seed_10876_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 02:40:22.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:22.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:35.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_11/index_5849_seed_10877_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:41:58.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 02:54:31.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_11/index_5747_seed_9374_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 02:55:54.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:54.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:08.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_11/index_5747_seed_9375_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:57:31.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 03:08:50.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_11/index_6297_seed_21130_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 03:10:16.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:10:16.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:27.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_11/index_6297_seed_21131_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:11:53.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 03:23:09.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:17.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_11/index_5262_seed_6554_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 03:24:33.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:33.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:25:46.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_11/index_5262_seed_6555_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:26:08.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 03:37:26.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:34.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_11/index_5667_seed_23636_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 03:38:51.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:51.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:07.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_11/index_5667_seed_23637_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:40:39.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 03:51:59.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_11/index_7042_seed_28226_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 03:53:29.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:29.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:54:38.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_11/index_7042_seed_28227_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:55:04.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:16.678 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 04:07:19.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:28.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_11/index_3228_seed_10884_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 04:08:44.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:44.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:01.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_11/index_3228_seed_10885_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:10:28.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 04:21:48.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_11/index_7092_seed_30148_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 04:23:14.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:23:14.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:28.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_11/index_7092_seed_30149_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 04:24:53.586 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:24:54.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 04:36:12.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:20.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_11/index_4111_seed_14044_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 04:37:38.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:38.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:38:50.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_11/index_4111_seed_14045_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:39:16.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 04:50:41.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:48.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_11/index_3987_seed_229_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 04:52:06.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:52:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:19.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_11/index_3987_seed_230_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:53:44.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 05:05:07.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:14.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_11/index_3203_seed_31848_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 05:06:30.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:30.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:07:43.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_11/index_3203_seed_31849_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:08:13.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 05:19:25.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:32.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_11/index_5774_seed_24444_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 05:20:46.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:46.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:21:58.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_11/index_5774_seed_24445_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:22:20.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 05:33:38.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:45.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_11/index_2273_seed_8959_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 05:35:03.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:35:03.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:15.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_11/index_2273_seed_8960_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:36:37.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 05:47:59.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_11/index_3420_seed_3325_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 05:49:26.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:26.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:35.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_11/index_3420_seed_3326_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 05:50:57.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:50:58.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 06:02:21.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:30.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_11/index_7752_seed_20266_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 06:03:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:49.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:04.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_11/index_7752_seed_20267_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:05:33.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 06:16:55.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_11/index_4774_seed_8906_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 06:18:20.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:20.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:31.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_11/index_4774_seed_8907_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:20:00.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:13.149 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 06:32:17.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:26.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_11/index_5559_seed_23233_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 06:33:42.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:42.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:34:55.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_11/index_5559_seed_23234_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:35:17.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:36.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 06:46:37.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:44.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_11/index_2093_seed_6487_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 06:48:01.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:48:01.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:15.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_11/index_2093_seed_6488_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:49:38.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 07:00:59.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:08.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_11/index_9851_seed_19316_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 07:02:26.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:26.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_11/index_9851_seed_19317_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:04:01.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 07:15:21.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:29.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_11/index_3980_seed_32064_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 07:16:57.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:57.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:06.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_11/index_3980_seed_32065_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:18:42.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 07:30:08.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:17.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_11/index_5769_seed_65_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 07:31:39.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:39.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:32:49.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_11/index_5769_seed_66_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:33:11.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.711 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 07:44:30.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:39.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_11/index_4123_seed_18850_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 07:45:57.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:57.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:04.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_11/index_4123_seed_18851_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:47:30.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 07:59:00.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:08.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_11/index_4959_seed_23113_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 08:00:25.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:25.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:01:41.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_11/index_4959_seed_23114_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:02:08.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 08:13:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_11/index_5905_seed_5558_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 08:14:59.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:59.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:17.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_11/index_5905_seed_5559_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:16:44.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.211 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 08:28:15.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_11/index_7831_seed_10104_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 08:29:44.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:44.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:30:56.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_11/index_7831_seed_10105_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:31:23.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 08:42:53.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:03.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_11/index_3044_seed_20188_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 08:44:20.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:20.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:45:35.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_11/index_3044_seed_20189_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:46:01.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:26.836 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 08:59:29.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:40.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_11/index_3100_seed_11388_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 09:01:04.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:01:04.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:11.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_11/index_3100_seed_11389_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 09:02:32.038 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:02:33.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 09:14:05.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:12.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_11/index_7069_seed_29312_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 09:15:43.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:43.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:16:52.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_11/index_7069_seed_29313_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:17:17.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 09:28:42.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_11/index_1218_seed_3912_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 09:30:09.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:30:09.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:27.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_11/index_1218_seed_3913_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:31:54.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 09:43:17.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:26.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_11/index_224_seed_4035_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 09:44:41.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:41.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:02.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_11/index_224_seed_4036_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 09:46:24.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:46:25.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:45.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 09:57:46.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:55.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_11/index_3058_seed_8421_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 09:59:16.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:59:16.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:24.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_11/index_3058_seed_8422_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:00:52.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 10:12:07.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:16.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_11/index_872_seed_4377_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 10:13:37.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:37.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:14:44.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_11/index_872_seed_4378_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:15:07.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 10:26:24.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:32.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_11/index_7663_seed_16228_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 10:27:55.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:55.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:02.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_11/index_7663_seed_16229_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:29:29.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:45.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 10:40:46.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_11/index_9007_seed_4148_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 10:42:19.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:42:19.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:28.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_11/index_9007_seed_4149_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:43:55.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 10:55:08.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_11/index_5512_seed_23078_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 10:56:36.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:36.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:57:53.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_11/index_5512_seed_23079_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:58:23.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 11:09:40.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_11/index_1433_seed_20242_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 11:11:09.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:11:09.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_11/index_1433_seed_20243_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:12:43.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:05.121 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 11:25:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:18.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_11/index_4205_seed_3892_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 11:26:34.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:34.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:27:53.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_11/index_4205_seed_3893_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:28:21.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 11:39:42.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:51.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_11/index_8388_seed_1442_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 11:41:12.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:41:12.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:19.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_11/index_8388_seed_1443_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:42:42.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 11:54:06.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:14.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_11/index_8771_seed_26396_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 11:55:31.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:31.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:56:42.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_11/index_8771_seed_26397_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:57:06.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 12:08:30.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:38.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_11/index_4087_seed_10873_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 12:09:53.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:53.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:07.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_11/index_4087_seed_10874_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:11:30.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 12:22:53.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_11/index_5572_seed_28435_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 12:24:21.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:21.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:30.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_11/index_5572_seed_28436_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:25:55.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 12:37:12.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_11/index_3035_seed_6463_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 12:38:38.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:38.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:39:47.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_11/index_3035_seed_6464_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:40:11.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 12:51:30.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:39.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_11/index_6815_seed_10107_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 12:53:04.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:53:04.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:11.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_11/index_6815_seed_10108_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:54:32.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:58.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 13:05:59.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:07.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_11/index_4533_seed_32617_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 13:07:22.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:22.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:08:38.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_11/index_4533_seed_32618_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:09:04.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 13:20:24.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:32.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_11/index_7087_seed_19901_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 13:21:47.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:47.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:00.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_11/index_7087_seed_19902_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:23:30.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 13:34:48.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:56.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_11/index_1304_seed_27786_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 13:36:11.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:36:11.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:25.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_11/index_1304_seed_27787_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:37:48.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.931 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 13:50:02.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_11/index_390_seed_31887_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 13:51:27.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:27.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:52:48.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_11/index_390_seed_31888_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:53:11.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 14:04:34.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:34.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:44.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_11/index_7867_seed_9808_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 14:06:06.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:06:06.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:19.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_11/index_7867_seed_9809_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:08:00.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 14:19:43.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:56.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_11/index_662_seed_3902_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 14:21:28.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:21:28.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:22:39.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_11/index_662_seed_3903_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 14:23:12.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:23:13.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 14:34:38.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:47.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_11/index_4046_seed_683_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 14:36:09.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:36:09.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:17.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_11/index_4046_seed_684_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:37:54.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:19.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 14:49:20.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_11/index_5175_seed_430_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 14:50:53.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:53.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:01.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_11/index_5175_seed_431_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:52:27.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 15:05:01.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:11.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_11/index_529_seed_2434_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 15:06:30.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:30.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:07:40.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_11/index_529_seed_2435_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 15:08:02.893 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:08:04.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:34.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 15:19:35.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:44.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_11/index_2592_seed_16888_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 15:21:05.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:21:05.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:15.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_11/index_2592_seed_16889_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:22:38.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 15:34:11.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_11/index_8289_seed_9134_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 15:35:41.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:41.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:36:53.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_11/index_8289_seed_9135_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:37:19.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.395 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 15:48:48.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:58.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_11/index_2781_seed_31925_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 15:50:16.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:50:16.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:26.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_11/index_2781_seed_31926_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 15:51:49.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:51:51.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 16:03:17.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:26.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_11/index_8988_seed_16010_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 16:04:46.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:46.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:05:57.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_11/index_8988_seed_16011_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:06:21.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:46.246 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 16:18:48.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:02.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_11/index_3580_seed_15400_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 16:20:19.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:19.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:21:34.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_11/index_3580_seed_15401_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:22:06.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 16:33:34.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:44.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_11/index_212_seed_16285_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 16:35:09.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:35:09.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:17.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_11/index_212_seed_16286_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 16:36:39.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:36:41.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 16:48:15.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:25.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_11/index_8900_seed_7680_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 16:49:46.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:46.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:50:54.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_11/index_8900_seed_7681_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:51:17.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 17:02:55.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:05.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_11/index_1332_seed_31991_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 17:04:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:23.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:34.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_11/index_1332_seed_31992_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:05:57.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 17:17:35.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:45.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_11/index_7930_seed_19479_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 17:19:06.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:19:06.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:16.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_11/index_7930_seed_19480_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:20:38.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 17:32:10.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:20.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_11/index_3526_seed_29621_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 17:33:40.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:40.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:01.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_11/index_3526_seed_29622_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:35:27.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.572 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 17:46:54.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:04.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_11/index_6236_seed_14252_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 17:48:24.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:24.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:49:39.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_11/index_6236_seed_14253_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 17:50:01.302 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:50:02.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:34.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 18:01:35.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:44.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_11/index_5829_seed_4882_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 18:03:05.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:03:05.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:18.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_11/index_5829_seed_4883_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 18:04:39.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:04:41.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 18:16:18.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:29.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_11/index_1373_seed_10995_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 18:17:50.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:50.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:01.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_11/index_1373_seed_10996_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:19:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 18:31:18.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:19.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:30.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_11/index_8211_seed_17346_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 18:33:00.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:33:00.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:15.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_11/index_8211_seed_17347_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:34:46.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:21.223 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 18:47:24.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:36.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_11/index_644_seed_10597_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 18:49:02.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:49:02.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:13.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_11/index_644_seed_10598_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:50:40.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 19:02:13.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:26.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_11/index_8611_seed_7696_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 19:03:46.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:46.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:01.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_11/index_8611_seed_7697_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:05:35.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 19:17:12.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:23.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_11/index_7550_seed_21776_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 19:18:44.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:44.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:19:59.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_11/index_7550_seed_21777_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:20:27.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 19:31:55.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:06.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_11/index_3273_seed_17555_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 19:33:30.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:30.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:34:55.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_11/index_3273_seed_17556_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:35:24.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:02.092 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 19:47:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:13.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_11/index_4105_seed_12589_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 19:48:35.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:35.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:49:48.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_11/index_4105_seed_12590_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:50:11.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:45.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 20:01:46.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:58.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_11/index_4418_seed_11488_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 20:03:16.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:03:16.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:29.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_11/index_4418_seed_11489_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 20:04:55.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:04:56.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 20:16:29.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:41.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_11/index_6525_seed_14263_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 20:18:06.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:18:06.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:19.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_11/index_6525_seed_14264_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:19:47.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 20:31:20.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:29.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_11/index_5824_seed_9811_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 20:32:52.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:52.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:00.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_11/index_5824_seed_9812_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:34:24.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:02.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 20:46:03.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:14.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_11/index_5150_seed_11842_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 20:47:36.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:36.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:48:50.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_11/index_5150_seed_11843_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:49:15.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 21:00:47.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:57.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_11/index_8824_seed_1346_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 21:02:23.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:02:23.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:32.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_11/index_8824_seed_1347_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:03:58.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:29.215 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 21:17:31.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:42.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_11/index_9585_seed_14340_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 21:19:02.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:19:02.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:13.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_11/index_9585_seed_14341_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:20:38.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 21:32:09.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:54:44.655 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-30 21:54:44.656 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-30 21:54:45.599 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:54:45.600 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 21:55:22.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:49.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_11/index_1871_seed_32875_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 21:57:12.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:57:12.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:18.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_11/index_1871_seed_32876_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 21:58:52.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:59:08.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.220 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 46, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 22:10:52.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:12:01.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_11/index_3841_seed_712_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 22:12:25.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:25.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:13:36.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_11/index_3841_seed_713_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:14:03.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 22:25:33.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:43.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_11/index_3109_seed_10553_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 22:27:00.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:27:01.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:15.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_11/index_3109_seed_10554_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:28:37.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 22:40:03.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:13.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_11/index_6414_seed_17502_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 22:41:34.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:34.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:42:46.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_11/index_6414_seed_17503_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 22:43:09.822 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:43:11.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 22:54:38.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:38.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:47.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_11/index_5259_seed_2698_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 22:56:08.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:56:08.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:26.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_11/index_5259_seed_2699_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 22:57:49.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:57:50.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:19.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 23:09:20.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:29.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_11/index_6402_seed_15149_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 23:11:01.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:11:01.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:11.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_11/index_6402_seed_15150_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:12:36.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 23:24:05.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:15.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_11/index_5821_seed_30564_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 23:25:37.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:37.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:26:47.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_11/index_5821_seed_30565_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:27:15.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.615 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 23:38:46.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:55.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_11/index_6055_seed_9340_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 23:40:14.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:40:14.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:25.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_11/index_6055_seed_9341_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:41:49.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:09.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-30 23:53:10.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:19.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_11/index_8903_seed_12168_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-30 23:54:38.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:38.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:55:53.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_11/index_8903_seed_12169_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:56:16.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 00:07:41.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_11/index_8562_seed_20162_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 00:09:13.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:09:13.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:25.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_11/index_8562_seed_20163_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:10:46.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.731 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:57.134 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 00:22:59.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_11/index_3672_seed_32787_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 00:24:30.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:30.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:25:41.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_11/index_3672_seed_32788_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:26:06.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 00:37:30.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:40.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_11/index_2179_seed_9286_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 00:39:01.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:39:01.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:18.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_11/index_2179_seed_9287_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:40:46.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 00:52:07.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:15.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_11/index_4971_seed_26688_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 00:53:36.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:36.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:54:49.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_11/index_4971_seed_26689_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:55:22.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 01:06:44.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:54.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_11/index_8020_seed_17272_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 01:08:15.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:08:15.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:27.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_11/index_8020_seed_17273_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:09:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 01:21:16.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:24.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_11/index_7420_seed_33081_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 01:22:44.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:44.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:23:56.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_11/index_7420_seed_33082_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:24:21.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 01:35:36.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:37.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_11/index_2730_seed_16331_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 01:37:03.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:37:03.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:16.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_11/index_2730_seed_16332_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:38:38.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 01:50:03.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:12.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_11/index_4503_seed_14774_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 01:51:32.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:32.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:52:42.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_11/index_4503_seed_14775_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 01:53:05.979 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:53:07.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.92s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:33.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 02:04:34.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:43.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_11/index_3263_seed_17977_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 02:06:00.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:06:00.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_11/index_3263_seed_17978_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:07:39.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 46, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-05-31 02:19:01.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:09.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_11/index_9452_seed_7775_dp_23_rank_46_sample_2_slot_0.mp4 +2026-05-31 02:20:31.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:31.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:21:39.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_11/index_9452_seed_7776_dp_23_rank_46_sample_3_slot_1.mp4 +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:22:03.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:30.692 | INFO | __main__:__init__:946 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-06-25 18:01:30.711 | INFO | __main__:__init__:947 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-06-25 18:01:31.385 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:31.385 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 18:04:09.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_11/index_1871_seed_32875_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 18:05:48.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:48.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:06:56.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_11/index_1871_seed_32876_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:07:32.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 46, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 18:18:41.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:47.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_11/index_3841_seed_712_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 18:20:04.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:20:04.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:12.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_11/index_3841_seed_713_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:21:35.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 18:32:26.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:32.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_11/index_3109_seed_10553_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 18:33:47.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:47.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:34:54.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_11/index_3109_seed_10554_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:35:15.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 18:46:08.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_11/index_6414_seed_17502_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 18:47:27.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:27.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:48:38.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_11/index_6414_seed_17503_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:49:03.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 19:00:02.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:09.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_11/index_5259_seed_2698_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 19:01:27.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:34.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_11/index_5259_seed_2699_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:02:57.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 19:13:49.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_11/index_6402_seed_15149_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 19:15:10.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:15:10.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:20.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_11/index_6402_seed_15150_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:16:42.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 19:27:32.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_11/index_5821_seed_30564_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 19:28:55.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:55.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:03.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_11/index_5821_seed_30565_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:30:24.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 19:41:17.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_11/index_6055_seed_9340_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 19:42:37.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:37.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:43:50.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_11/index_6055_seed_9341_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:44:15.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 19:55:08.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_11/index_8903_seed_12168_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 19:56:29.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:29.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:57:41.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_11/index_8903_seed_12169_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:58:03.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 20:09:02.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:08.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_11/index_8562_seed_20162_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 20:10:22.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:22.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:35.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_11/index_8562_seed_20163_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:12:00.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.949 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 20:23:32.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:40.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_11/index_3672_seed_32787_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 20:24:57.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:57.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_11/index_3672_seed_32788_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:26:26.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 20:37:18.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:19.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_11/index_2179_seed_9286_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 20:38:39.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:39.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:39:48.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_11/index_2179_seed_9287_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:40:09.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 20:51:03.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:09.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_11/index_4971_seed_26688_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 20:52:24.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:24.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:35.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_11/index_4971_seed_26689_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:53:57.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 21:04:50.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_11/index_8020_seed_17272_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 21:06:13.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:06:13.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:23.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_11/index_8020_seed_17273_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:07:49.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 21:18:43.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:49.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_11/index_7420_seed_33081_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 21:20:05.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:20:05.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:17.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_11/index_7420_seed_33082_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:21:39.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 21:32:31.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_11/index_2730_seed_16331_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 21:33:54.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:54.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:01.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_11/index_2730_seed_16332_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:35:21.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.656 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 21:46:16.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:22.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_11/index_4503_seed_14774_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 21:47:38.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:38.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:48:47.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_11/index_4503_seed_14775_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:49:10.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 22:00:05.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_11/index_3263_seed_17977_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 22:01:26.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:26.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:37.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_11/index_3263_seed_17978_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:02:58.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 22:13:50.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_11/index_9452_seed_7775_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 22:15:11.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:15:11.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:22.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_11/index_9452_seed_7776_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:16:45.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 22:27:39.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:45.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_11/index_1765_seed_27681_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 22:29:00.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:29:00.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:11.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_11/index_1765_seed_27682_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:30:32.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:58.183 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 22:41:59.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_11/index_7039_seed_23291_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 22:43:24.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:24.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:31.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_11/index_7039_seed_23292_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 22:44:51.321 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:44:52.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.064 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 22:55:45.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:51.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_11/index_856_seed_10285_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 22:57:10.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:57:10.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:17.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_11/index_856_seed_10286_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 22:58:38.406 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:58:39.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 23:09:30.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_11/index_9375_seed_29287_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 23:10:54.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:54.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:01.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_11/index_9375_seed_29288_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:12:23.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.220 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 23:23:15.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_11/index_5032_seed_22409_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 23:24:39.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:39.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:25:47.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_11/index_5032_seed_22410_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:26:11.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 23:37:07.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:13.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_11/index_5849_seed_10876_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 23:38:31.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:31.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:39:38.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_11/index_5849_seed_10877_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:40:02.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-25 23:51:52.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:58.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_11/index_5747_seed_9374_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-25 23:53:13.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:53:13.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:23.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_11/index_5747_seed_9375_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:54:43.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 00:05:36.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:42.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_11/index_6297_seed_21130_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 00:06:55.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:55.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:07.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_11/index_6297_seed_21131_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:08:28.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 00:19:21.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_11/index_5262_seed_6554_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 00:20:41.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:41.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:21:50.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_11/index_5262_seed_6555_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:22:11.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 00:33:07.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:13.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_11/index_5667_seed_23636_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 00:34:27.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:27.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:35:37.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_11/index_5667_seed_23637_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:36:04.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 00:47:03.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:09.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_11/index_7042_seed_28226_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 00:48:27.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:27.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:38.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_11/index_7042_seed_28227_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:50:00.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:28.735 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 01:01:30.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:31.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_11/index_3228_seed_10884_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 01:02:53.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:53.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:03.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_11/index_3228_seed_10885_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:04:25.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 01:15:17.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_11/index_7092_seed_30148_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 01:16:37.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:37.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:17:50.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_11/index_7092_seed_30149_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:18:11.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 01:29:08.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:14.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_11/index_4111_seed_14044_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 01:30:29.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:29.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:31:40.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_11/index_4111_seed_14045_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:32:03.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 01:42:58.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_11/index_3987_seed_229_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 01:44:19.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:19.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:29.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_11/index_3987_seed_230_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:45:50.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 01:56:44.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_11/index_3203_seed_31848_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 01:58:05.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:58:05.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:17.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_11/index_3203_seed_31849_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:59:40.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 02:10:30.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:37.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_11/index_5774_seed_24444_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 02:11:50.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:50.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:03.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_11/index_5774_seed_24445_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 02:13:23.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:13:24.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 02:24:16.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_11/index_2273_seed_8959_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 02:25:36.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:36.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:26:46.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_11/index_2273_seed_8960_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:27:11.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.262 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 02:38:08.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_11/index_3420_seed_3325_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 02:39:29.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:29.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_11/index_3420_seed_3326_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:41:00.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 02:51:53.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:54.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_11/index_7752_seed_20266_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 02:53:13.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:53:13.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:25.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_11/index_7752_seed_20267_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:54:48.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 03:05:40.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_11/index_4774_seed_8906_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 03:07:01.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:07:01.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:15.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_11/index_4774_seed_8907_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:08:38.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:05.107 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 03:20:06.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:14.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_11/index_5559_seed_23233_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 03:21:28.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:28.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:39.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_11/index_5559_seed_23234_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:22:59.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 03:33:52.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_11/index_2093_seed_6487_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 03:35:13.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:35:13.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:23.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_11/index_2093_seed_6488_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:36:46.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 03:47:40.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_11/index_9851_seed_19316_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 03:48:59.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:59.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:10.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_11/index_9851_seed_19317_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:50:31.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 04:01:21.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:27.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_11/index_3980_seed_32064_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 04:02:46.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:46.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:03:52.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_11/index_3980_seed_32065_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:04:12.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 04:15:09.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_11/index_5769_seed_65_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 04:16:35.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:35.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:17:42.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_11/index_5769_seed_66_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:18:05.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 04:29:03.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:09.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_11/index_4123_seed_18850_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 04:30:23.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:23.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:33.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_11/index_4123_seed_18851_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:31:54.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 04:42:46.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:52.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_11/index_4959_seed_23113_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 04:44:04.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:44:04.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:16.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_11/index_4959_seed_23114_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:45:35.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.442 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 04:56:27.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_11/index_5905_seed_5558_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 04:57:48.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:48.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:58:58.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_11/index_5905_seed_5559_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:59:19.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 05:10:10.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:16.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_11/index_7831_seed_10104_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 05:11:35.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:35.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:12:41.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_11/index_7831_seed_10105_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:13:03.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 05:24:03.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_11/index_3044_seed_20188_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 05:25:24.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:24.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:35.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_11/index_3044_seed_20189_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:26:58.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.108 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 05:39:29.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_11/index_3100_seed_11388_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 05:40:50.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:50.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:00.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_11/index_3100_seed_11389_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:42:20.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 05:53:10.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:16.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_11/index_7069_seed_29312_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 05:54:33.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:33.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:39.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_11/index_7069_seed_29313_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:56:00.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.549 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 06:06:52.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:58.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_11/index_1218_seed_3912_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 06:08:13.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:08:13.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:22.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_11/index_1218_seed_3913_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:09:44.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 06:20:40.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_11/index_224_seed_4035_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 06:22:01.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:22:01.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:09.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_11/index_224_seed_4036_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:23:31.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 06:34:21.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:22.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_11/index_3058_seed_8421_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 06:35:42.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:42.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:36:51.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_11/index_3058_seed_8422_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:37:11.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 06:48:09.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:15.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_11/index_872_seed_4377_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 06:49:30.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:30.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:50:44.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_11/index_872_seed_4378_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:51:03.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 07:02:03.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_11/index_7663_seed_16228_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 07:03:29.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:29.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:34.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_11/index_7663_seed_16229_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:04:55.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 07:15:48.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:54.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_11/index_9007_seed_4148_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 07:17:16.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:17:16.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:21.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_11/index_9007_seed_4149_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:18:41.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 07:29:32.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_11/index_5512_seed_23078_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 07:30:53.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:53.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:04.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_11/index_5512_seed_23079_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:32:31.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 07:43:21.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:27.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_11/index_1433_seed_20242_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 07:44:41.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:41.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:45:53.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_11/index_1433_seed_20243_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:46:15.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:44.144 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 07:57:45.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:53.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_11/index_4205_seed_3892_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 07:59:07.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:59:07.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:18.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_11/index_4205_seed_3893_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:00:40.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 08:11:33.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:39.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_11/index_8388_seed_1442_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 08:12:53.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:53.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:03.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_11/index_8388_seed_1443_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:14:26.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 08:25:18.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:24.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_11/index_8771_seed_26396_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 08:26:39.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:39.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:27:48.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_11/index_8771_seed_26397_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:28:07.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 08:39:03.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_11/index_4087_seed_10873_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 08:40:24.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:24.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:36.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_11/index_4087_seed_10874_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:41:57.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 08:52:50.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_11/index_5572_seed_28435_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 08:54:10.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:54:10.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:22.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_11/index_5572_seed_28436_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:55:45.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.550 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 09:06:39.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:45.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_11/index_3035_seed_6463_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 09:08:00.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:08:00.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:11.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_11/index_3035_seed_6464_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:09:33.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 09:20:22.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:29.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_11/index_6815_seed_10107_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 09:21:42.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:42.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:22:52.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_11/index_6815_seed_10108_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:23:12.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 09:34:09.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:15.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_11/index_4533_seed_32617_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 09:35:28.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:28.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:36:41.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_11/index_4533_seed_32618_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:37:04.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 09:47:56.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_11/index_7087_seed_19901_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 09:49:15.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:15.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:29.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_11/index_7087_seed_19902_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:50:51.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 10:01:45.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_11/index_1304_seed_27786_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 10:03:04.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:03:04.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:15.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_11/index_1304_seed_27787_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 10:04:34.031 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:04:34.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.855 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 10:16:01.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_11/index_390_seed_31887_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 10:17:24.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:24.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:34.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_11/index_390_seed_31888_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 10:18:58.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:18:59.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 10:29:52.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:58.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_11/index_7867_seed_9808_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 10:31:13.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:31:13.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:23.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_11/index_7867_seed_9809_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:32:44.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 10:43:37.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:43.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_11/index_662_seed_3902_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 10:44:57.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:57.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:08.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_11/index_662_seed_3903_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:46:29.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 10:57:21.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:27.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_11/index_4046_seed_683_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 10:58:42.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:42.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:59:56.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_11/index_4046_seed_684_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:00:18.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 11:11:08.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:09.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_11/index_5175_seed_430_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 11:12:32.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:32.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:39.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_11/index_5175_seed_431_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:13:59.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 11:26:08.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:14.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_11/index_529_seed_2434_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 11:27:27.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:27.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:28:40.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_11/index_529_seed_2435_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 11:29:02.143 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:29:03.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 11:39:55.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_11/index_2592_seed_16888_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 11:41:19.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:19.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:24.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_11/index_2592_seed_16889_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:42:44.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 11:53:37.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_11/index_8289_seed_9134_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 11:54:53.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:53.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:07.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_11/index_8289_seed_9135_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:56:31.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 12:07:21.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:27.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_11/index_2781_seed_31925_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 12:08:43.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:43.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:09:51.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_11/index_2781_seed_31926_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:10:12.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 12:21:06.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_11/index_8988_seed_16010_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 12:22:27.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:27.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:39.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_11/index_8988_seed_16011_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 12:23:59.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:24:00.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.790 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 12:35:30.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:31.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:38.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_11/index_3580_seed_15400_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 12:36:55.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:55.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:04.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_11/index_3580_seed_15401_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:38:25.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 12:49:16.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:22.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_11/index_212_seed_16285_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 12:50:38.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:38.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:51:48.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_11/index_212_seed_16286_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 12:52:08.745 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:52:09.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 13:03:03.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_11/index_8900_seed_7680_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 13:04:27.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:27.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:33.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_11/index_8900_seed_7681_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:05:53.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 13:16:44.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_11/index_1332_seed_31991_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 13:18:04.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:18:04.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:16.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_11/index_1332_seed_31992_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:19:41.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 13:30:31.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_11/index_7930_seed_19479_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 13:31:53.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:53.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:05.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_11/index_7930_seed_19480_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:33:25.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 13:44:15.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:21.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_11/index_3526_seed_29621_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 13:45:39.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:39.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:46:49.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_11/index_3526_seed_29622_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 13:47:12.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:47:13.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 13:58:07.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:13.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_11/index_6236_seed_14252_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 13:59:28.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:28.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:36.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_11/index_6236_seed_14253_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:00:56.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 14:11:49.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_11/index_5829_seed_4882_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 14:13:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:13:10.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:18.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_11/index_5829_seed_4883_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 14:14:38.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:14:39.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 14:25:28.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_11/index_1373_seed_10995_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 14:26:47.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:47.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:02.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_11/index_1373_seed_10996_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:28:25.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 14:39:15.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_11/index_8211_seed_17346_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 14:40:43.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:43.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:41:50.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_11/index_8211_seed_17347_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:42:13.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:43.103 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 14:53:45.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_11/index_644_seed_10597_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 14:55:12.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:55:12.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:16.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_11/index_644_seed_10598_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:56:36.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 15:07:27.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_11/index_8611_seed_7696_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 15:08:47.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:47.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:09:56.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_11/index_8611_seed_7697_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:10:19.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 15:21:11.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_11/index_7550_seed_21776_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 15:22:33.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:33.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:23:41.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_11/index_7550_seed_21777_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:24:02.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 15:34:56.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_11/index_3273_seed_17555_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 15:36:25.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:25.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:30.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_11/index_3273_seed_17556_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:37:50.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 15:48:45.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_11/index_4105_seed_12589_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 15:50:05.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:50:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:15.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_11/index_4105_seed_12590_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:51:35.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 16:02:26.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_11/index_4418_seed_11488_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 16:03:47.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:47.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:04:57.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_11/index_4418_seed_11489_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:05:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.281 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 16:16:06.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_11/index_6525_seed_14263_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 16:17:29.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:29.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:18:42.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_11/index_6525_seed_14264_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:19:04.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 16:29:54.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_11/index_5824_seed_9811_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 16:31:14.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:14.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:24.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_11/index_5824_seed_9812_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 16:32:44.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:32:45.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 16:43:38.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:44.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_11/index_5150_seed_11842_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 16:45:01.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:45:01.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:08.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_11/index_5150_seed_11843_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:46:31.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 16:57:20.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:21.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_11/index_8824_seed_1346_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 16:58:48.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:48.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:59:54.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_11/index_8824_seed_1347_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:00:16.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:50.837 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 17:12:53.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_11/index_9585_seed_14340_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 17:14:16.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:14:16.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:26.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_11/index_9585_seed_14341_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 17:15:47.056 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:15:48.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 17:26:39.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:45.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_11/index_732_seed_15323_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 17:28:03.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:28:03.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:09.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_11/index_732_seed_15324_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:29:30.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 17:40:20.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:26.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_11/index_6074_seed_3270_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 17:41:41.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:41.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:42:54.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_11/index_6074_seed_3271_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:43:14.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 17:54:06.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:12.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_11/index_6252_seed_8029_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 17:55:32.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:32.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:38.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_11/index_6252_seed_8030_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 17:56:57.969 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:56:59.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 18:07:49.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_11/index_2448_seed_10973_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 18:09:11.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:09:11.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:21.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_11/index_2448_seed_10974_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:10:45.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 18:21:38.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:44.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_11/index_4795_seed_27425_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 18:22:59.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:59.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:12.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_11/index_4795_seed_27426_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 18:24:36.191 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:24:37.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 18:35:27.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:28.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_11/index_5700_seed_13740_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 18:36:53.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:53.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:03.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_11/index_5700_seed_13741_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:38:24.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 18:49:15.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_11/index_5422_seed_22565_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 18:50:38.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:38.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:51:45.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_11/index_5422_seed_22566_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:52:05.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 19:03:01.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_11/index_5054_seed_18300_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 19:04:24.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:24.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:35.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_11/index_5054_seed_18301_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:05:55.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 19:16:46.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:52.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_11/index_4956_seed_29262_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 19:18:06.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:18:06.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_11/index_4956_seed_29263_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:19:40.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.801 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 19:31:08.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_11/index_8549_seed_21814_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 19:32:38.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:38.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:33:43.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_11/index_8549_seed_21815_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 19:34:06.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:34:07.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.122 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 19:45:03.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_11/index_7082_seed_5586_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 19:46:22.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:22.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:32.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_11/index_7082_seed_5587_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:47:52.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 19:58:45.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:45.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_11/index_3331_seed_24676_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 20:00:07.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:00:07.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:17.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_11/index_3331_seed_24677_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:01:38.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 20:12:28.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_11/index_6102_seed_32885_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 20:13:52.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:52.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:14:59.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_11/index_6102_seed_32886_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:15:22.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 20:26:14.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_11/index_2023_seed_14139_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 20:27:34.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:34.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:28:48.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_11/index_2023_seed_14140_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:29:12.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 20:40:07.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_11/index_2874_seed_6242_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 20:41:27.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:27.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:42:38.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_11/index_2874_seed_6243_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:43:01.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 20:53:52.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_11/index_5799_seed_31077_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 20:55:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:55:16.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:25.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_11/index_5799_seed_31078_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:56:54.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 21:07:44.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:50.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_11/index_138_seed_6173_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 21:09:04.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:09:04.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:18.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_11/index_138_seed_6174_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:10:42.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 21:21:31.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_11/index_7837_seed_18150_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 21:22:52.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:52.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:02.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_11/index_7837_seed_18151_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:24:26.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 21:35:16.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_11/index_9175_seed_6666_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 21:36:41.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:41.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:37:49.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_11/index_9175_seed_6667_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:38:11.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.882 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 21:49:43.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:51.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_11/index_6893_seed_7310_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 21:51:10.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:51:10.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:15.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_11/index_6893_seed_7311_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:52:38.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 22:03:26.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_11/index_1215_seed_22726_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 22:04:47.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:47.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:05:56.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_11/index_1215_seed_22727_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:06:18.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.373 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 22:17:07.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:13.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_11/index_2708_seed_20057_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 22:18:27.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:27.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:37.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_11/index_2708_seed_20058_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 22:19:58.738 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:19:59.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 22:30:49.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_11/index_7140_seed_22007_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 22:32:11.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:32:11.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:20.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_11/index_7140_seed_22008_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:33:43.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 22:44:32.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:38.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_11/index_1256_seed_11341_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 22:45:53.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:53.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:04.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_11/index_1256_seed_11342_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:47:24.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 22:59:16.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:22.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_11/index_4246_seed_24473_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 23:00:35.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:35.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:01:48.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_11/index_4246_seed_24474_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:02:09.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 23:13:02.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_11/index_1206_seed_4628_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 23:14:21.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:21.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:32.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_11/index_1206_seed_4629_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:15:53.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 23:26:44.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:50.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_11/index_472_seed_29702_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 23:28:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:28:07.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_11/index_472_seed_29703_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:29:39.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 23:40:27.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:33.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_11/index_4780_seed_28154_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 23:41:48.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:48.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:01.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_11/index_4780_seed_28155_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:43:26.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-26 23:54:16.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_11/index_6698_seed_13595_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-26 23:55:37.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:37.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:56:46.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_11/index_6698_seed_13596_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:57:09.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.630 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 00:08:40.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:48.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_11/index_5683_seed_24095_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 00:10:04.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:10:04.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:11.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_11/index_5683_seed_24096_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:11:32.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 00:22:20.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:26.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_11/index_6739_seed_4791_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 00:23:46.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:46.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:24:53.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_11/index_6739_seed_4792_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:25:15.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 00:36:06.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:12.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_11/index_8952_seed_33384_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 00:37:26.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:26.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:38.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_11/index_8952_seed_33385_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:38:59.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 00:49:49.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_11/index_3169_seed_27040_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 00:51:08.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:51:08.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:19.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_11/index_3169_seed_27041_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:52:44.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 01:03:35.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_11/index_8920_seed_23987_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 01:04:56.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:56.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:05.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_11/index_8920_seed_23988_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:06:28.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 01:17:20.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:20.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_11/index_5444_seed_24998_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 01:18:46.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:46.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:19:52.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_11/index_5444_seed_24999_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:20:15.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 01:31:07.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_11/index_7664_seed_23890_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 01:32:26.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:26.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:38.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_11/index_7664_seed_23891_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:33:58.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 01:44:47.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_11/index_8484_seed_18786_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 01:46:05.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:46:05.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:19.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_11/index_8484_seed_18787_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:47:40.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 01:58:27.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_11/index_2693_seed_34_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 01:59:48.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:48.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:01.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_11/index_2693_seed_35_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 02:01:22.441 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:01:23.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 02:12:14.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:15.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:21.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_11/index_4907_seed_5365_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 02:13:35.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:35.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:14:47.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_11/index_4907_seed_5366_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 02:15:08.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:15:09.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:36.943 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 02:26:39.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_11/index_6814_seed_1989_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 02:28:02.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:28:02.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:13.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_11/index_6814_seed_1990_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:29:34.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 02:40:23.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:29.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_11/index_310_seed_18790_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 02:41:44.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:44.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:00.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_11/index_310_seed_18791_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:43:25.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 02:54:14.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:20.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_11/index_6330_seed_2302_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 02:55:32.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:32.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:56:47.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_11/index_6330_seed_2303_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:57:09.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 03:08:01.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:02.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_11/index_7141_seed_19172_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 03:09:22.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:22.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:31.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_11/index_7141_seed_19173_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:10:50.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 03:21:41.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:47.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_11/index_907_seed_22671_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 03:23:06.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:23:06.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:15.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_11/index_907_seed_22672_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:24:38.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 03:35:26.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:32.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_11/index_1795_seed_21859_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 03:36:48.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:48.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:02.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_11/index_1795_seed_21860_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:38:25.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 03:49:14.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:20.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_11/index_6470_seed_18970_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 03:50:39.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:39.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:51:45.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_11/index_6470_seed_18971_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:52:08.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 04:03:03.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:09.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_11/index_7381_seed_18872_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 04:04:23.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:23.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:33.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_11/index_7381_seed_18873_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:05:54.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 04:16:48.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_11/index_7347_seed_30399_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 04:18:06.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:18:06.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:20.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_11/index_7347_seed_30400_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:19:42.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 04:30:31.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_11/index_6241_seed_23797_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 04:31:51.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:51.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:01.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_11/index_6241_seed_23798_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:33:22.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.416 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 04:45:58.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_11/index_5587_seed_26670_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 04:47:19.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:19.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:31.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_11/index_5587_seed_26671_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:48:56.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 04:59:46.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_11/index_5889_seed_26223_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 05:01:07.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:01:07.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:17.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_11/index_5889_seed_26224_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:02:40.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 05:13:28.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:33.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_11/index_1756_seed_15318_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 05:14:49.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:49.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:15:56.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_11/index_1756_seed_15319_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:16:17.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 05:27:07.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:08.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_11/index_177_seed_15353_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 05:28:28.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:29:40.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_11/index_177_seed_15354_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:30:01.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 05:40:50.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:56.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_11/index_5730_seed_7288_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 05:42:11.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:42:11.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:19.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_11/index_5730_seed_7289_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:43:44.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 05:54:35.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_11/index_881_seed_24977_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 05:55:58.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:58.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:07.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_11/index_881_seed_24978_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:57:30.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 06:08:19.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_11/index_5590_seed_4452_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 06:09:40.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:40.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:10:50.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_11/index_5590_seed_4453_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:11:14.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 06:22:07.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_11/index_888_seed_24147_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 06:23:28.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:28.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:38.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_11/index_888_seed_24148_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 06:24:58.738 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:24:59.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 06:35:50.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_11/index_6353_seed_8901_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 06:37:11.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:37:11.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:20.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_11/index_6353_seed_8902_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:38:42.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 06:49:31.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:36.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_11/index_7978_seed_14044_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 06:50:52.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:52.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:02.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_11/index_7978_seed_14045_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:52:21.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:48.140 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 07:03:49.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_11/index_4715_seed_16561_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 07:05:10.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:05:10.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:20.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_11/index_4715_seed_16562_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:06:40.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 07:17:29.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_11/index_9180_seed_15719_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 07:18:50.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:50.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:00.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_11/index_9180_seed_15720_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 07:20:21.813 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:20:23.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 07:31:16.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_11/index_2992_seed_7557_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 07:32:38.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:38.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:33:49.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_11/index_2992_seed_7558_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:34:12.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 07:45:07.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:13.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_11/index_3048_seed_10582_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 07:46:28.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:28.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:47:38.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_11/index_3048_seed_10583_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:48:02.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 07:58:52.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_11/index_8691_seed_16690_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 08:00:16.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:00:16.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:23.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_11/index_8691_seed_16691_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:01:46.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 08:12:38.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_11/index_3581_seed_27736_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 08:14:00.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:14:00.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:10.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_11/index_3581_seed_27737_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:15:31.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 08:26:20.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:26.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_11/index_9235_seed_5697_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 08:27:41.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:41.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:28:51.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_11/index_9235_seed_5698_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:29:11.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.142 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 08:40:03.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_11/index_8739_seed_3765_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 08:41:29.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:29.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:35.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_11/index_8739_seed_3766_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:42:57.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 08:53:47.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_11/index_1852_seed_5757_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 08:55:08.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:55:08.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:23.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_11/index_1852_seed_5758_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:56:50.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 09:07:41.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:47.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_11/index_5236_seed_7693_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 09:09:01.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:09:01.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:13.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_11/index_5236_seed_7694_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:10:37.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:00.730 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 09:22:02.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:02.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_11/index_3592_seed_12277_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 09:23:21.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:21.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:34.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_11/index_3592_seed_12278_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:24:55.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 09:35:43.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_11/index_7926_seed_31287_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 09:37:01.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:37:01.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:13.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_11/index_7926_seed_31288_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 09:38:34.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:38:35.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 09:49:26.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_11/index_6964_seed_14391_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 09:50:48.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:48.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:51:57.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_11/index_6964_seed_14392_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:52:17.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 10:03:08.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_11/index_5720_seed_5028_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 10:04:30.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:30.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:39.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_11/index_5720_seed_5029_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 10:05:59.863 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:06:00.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 10:16:53.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_11/index_5115_seed_6796_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 10:18:14.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:18:14.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:24.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_11/index_5115_seed_6797_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:19:46.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 10:31:49.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_11/index_1557_seed_8347_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 10:33:11.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:33:11.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:21.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_11/index_1557_seed_8348_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:34:43.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 10:45:31.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:32.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_11/index_3256_seed_23979_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 10:46:53.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:53.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:03.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_11/index_3256_seed_23980_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:48:27.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 10:59:16.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_11/index_2731_seed_31289_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 11:00:37.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:37.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:01:49.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_11/index_2731_seed_31290_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:02:13.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.909 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.246 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 11:13:07.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_11/index_2803_seed_15597_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 11:14:29.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:29.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:35.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_11/index_2803_seed_15598_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:15:55.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 11:26:47.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:53.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_11/index_1991_seed_32292_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 11:28:11.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:28:11.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:21.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_11/index_1991_seed_32293_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:29:42.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.004 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 11:41:06.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_11/index_8352_seed_14925_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 11:42:28.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:28.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:38.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_11/index_8352_seed_14926_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 11:43:58.131 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:43:59.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 11:54:48.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_11/index_2171_seed_29704_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 11:56:15.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:56:15.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:19.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_11/index_2171_seed_29705_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:57:45.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 12:08:34.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_11/index_6467_seed_20490_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 12:09:59.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:59.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:05.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_11/index_6467_seed_20491_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:11:27.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 12:22:19.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:25.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_11/index_9667_seed_15242_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 12:23:42.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:42.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:24:50.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_11/index_9667_seed_15243_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:25:10.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 12:36:01.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:07.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_11/index_2523_seed_24985_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 12:37:28.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:28.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:33.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_11/index_2523_seed_24986_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 12:38:52.462 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:38:53.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 12:49:44.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_11/index_8432_seed_16324_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 12:51:06.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:51:06.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:14.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_11/index_8432_seed_16325_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:52:35.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 13:03:21.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_11/index_8884_seed_8465_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 13:04:41.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:41.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:05:51.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_11/index_8884_seed_8466_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:06:13.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 13:17:08.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_11/index_7521_seed_19142_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 13:18:27.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:27.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:37.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_11/index_7521_seed_19143_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:19:58.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 13:30:47.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_11/index_155_seed_634_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 13:32:06.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:32:06.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:21.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_11/index_155_seed_635_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:33:47.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 13:44:38.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:44.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_11/index_3834_seed_10128_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 13:45:59.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:59.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:09.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_11/index_3834_seed_10129_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:47:31.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:54.989 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 13:58:57.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:03.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_11/index_9249_seed_12155_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 14:00:17.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:17.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:30.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_11/index_9249_seed_12156_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:01:51.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 14:12:43.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:49.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_11/index_595_seed_14041_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 14:14:03.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:14:03.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:12.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_11/index_595_seed_14042_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 14:15:31.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:15:32.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 14:26:20.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_11/index_8174_seed_28401_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 14:27:42.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:42.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:28:50.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_11/index_8174_seed_28402_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:29:12.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:05.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 14:40:05.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:06.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_11/index_480_seed_25653_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 14:41:26.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:26.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:36.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_11/index_480_seed_25654_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:42:58.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 14:53:47.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_11/index_5608_seed_22575_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 14:55:09.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:55:09.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:18.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_11/index_5608_seed_22576_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:56:39.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 15:07:26.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_11/index_6533_seed_27742_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 15:08:48.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:48.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:09:56.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_11/index_6533_seed_27743_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:10:18.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 15:21:06.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:12.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_11/index_108_seed_32987_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 15:22:26.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:26.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:37.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_11/index_108_seed_32988_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:23:58.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 15:34:47.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:47.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_11/index_8563_seed_28299_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 15:36:10.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:36:10.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:20.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_11/index_8563_seed_28300_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:37:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.331 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 15:48:32.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_11/index_5042_seed_7742_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 15:49:57.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:57.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:03.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_11/index_5042_seed_7743_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:51:25.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 16:02:14.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:20.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_11/index_98_seed_12928_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 16:03:40.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:40.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:04:44.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_11/index_98_seed_12929_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:05:05.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.635 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 16:17:43.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_11/index_8228_seed_32526_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 16:19:06.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:19:06.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:17.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_11/index_8228_seed_32527_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:20:37.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 16:31:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_11/index_3896_seed_14579_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 16:32:48.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:48.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:33:58.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_11/index_3896_seed_14580_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:34:18.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 16:45:07.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:13.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_11/index_7640_seed_30333_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 16:46:28.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:28.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:38.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_11/index_7640_seed_30334_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:47:59.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 16:58:47.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:53.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_11/index_1602_seed_5217_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 17:00:11.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:00:11.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:20.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_11/index_1602_seed_5218_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:01:42.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 17:12:31.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_11/index_9783_seed_21755_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 17:13:54.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:54.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:02.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_11/index_9783_seed_21756_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:15:22.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 17:26:15.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:21.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_11/index_8448_seed_5562_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 17:27:35.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:35.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:28:46.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_11/index_8448_seed_5563_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 17:29:07.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:29:09.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 17:40:00.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:06.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_11/index_7198_seed_23854_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 17:41:23.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:23.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:34.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_11/index_7198_seed_23855_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:42:55.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 17:53:45.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_11/index_2913_seed_4963_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 17:55:10.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:55:10.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:15.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_11/index_2913_seed_4964_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:56:38.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 18:07:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:31.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_11/index_9362_seed_11500_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 18:08:48.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:48.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:01.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_11/index_9362_seed_11501_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:10:24.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 18:21:14.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_11/index_796_seed_8653_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 18:22:39.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:39.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:23:45.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_11/index_796_seed_8654_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:24:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.050 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 18:35:38.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:44.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_11/index_9830_seed_1658_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 18:36:58.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:58.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:09.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_11/index_9830_seed_1659_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:38:32.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 18:49:21.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_11/index_7147_seed_22677_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 18:50:42.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:43.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:51:55.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_11/index_7147_seed_22678_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:52:16.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 19:03:06.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:12.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_11/index_2525_seed_24639_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 19:04:28.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:28.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:36.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_11/index_2525_seed_24640_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:05:55.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 19:16:46.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_11/index_739_seed_1163_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 19:18:08.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:18:08.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:21.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_11/index_739_seed_1164_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:19:42.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 19:30:30.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:36.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_11/index_4785_seed_22630_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 19:31:55.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:55.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:04.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_11/index_4785_seed_22631_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:33:25.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 19:44:14.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_11/index_2024_seed_32228_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 19:45:39.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:39.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:46:44.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_11/index_2024_seed_32229_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:47:10.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 19:58:02.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:08.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_11/index_6719_seed_10594_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 19:59:23.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:23.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:33.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_11/index_6719_seed_10595_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:00:54.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 20:11:43.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_11/index_7403_seed_10688_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 20:13:05.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:13:05.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:15.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_11/index_7403_seed_10689_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:14:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 20:25:26.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_11/index_1540_seed_12675_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 20:26:54.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:54.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:27:58.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_11/index_1540_seed_12676_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 20:28:17.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:28:19.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 20:39:06.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:13.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_11/index_4394_seed_28717_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 20:40:30.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:30.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:41:38.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_11/index_4394_seed_28718_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 20:42:03.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:42:04.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.708 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 20:53:29.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:35.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_11/index_6646_seed_28080_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 20:54:54.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:54.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:55:59.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_11/index_6646_seed_28081_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 20:56:20.307 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:56:21.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 21:07:10.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:16.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_11/index_5878_seed_27435_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 21:08:32.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:32.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:09:41.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_11/index_5878_seed_27436_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:10:07.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 21:21:01.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_11/index_5401_seed_3051_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 21:22:24.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:24.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:31.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_11/index_5401_seed_3052_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:23:52.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 21:34:42.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:48.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_11/index_7932_seed_20740_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 21:36:03.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:36:03.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:12.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_11/index_7932_seed_20741_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 21:37:33.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:37:34.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 21:48:24.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:30.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_11/index_6725_seed_5074_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 21:49:45.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:45.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:50:54.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_11/index_6725_seed_5075_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:51:15.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 22:03:21.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:27.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_11/index_9504_seed_20896_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 22:04:41.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:41.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:05:54.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_11/index_9504_seed_20897_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:06:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 22:17:06.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:12.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_11/index_7907_seed_20268_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 22:18:27.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:27.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:37.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_11/index_7907_seed_20269_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:19:57.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 22:30:47.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:53.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_11/index_6842_seed_15050_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 22:32:06.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:32:06.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:21.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_11/index_6842_seed_15051_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:33:41.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 22:44:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:36.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_11/index_3633_seed_19639_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 22:45:51.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:51.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:46:59.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_11/index_3633_seed_19640_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:47:18.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 22:58:07.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_11/index_8260_seed_33310_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 22:59:28.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:28.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:00:38.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_11/index_8260_seed_33311_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:01:01.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.859 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 23:12:29.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_11/index_4729_seed_2858_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 23:13:53.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:53.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:01.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_11/index_4729_seed_2859_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:15:24.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 23:26:15.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_11/index_3060_seed_3748_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 23:27:36.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:36.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:28:45.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_11/index_3060_seed_3749_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 23:29:05.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:29:06.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 23:40:02.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:08.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_11/index_9259_seed_28726_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 23:41:22.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:22.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:42:36.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_11/index_9259_seed_28727_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:43:01.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-27 23:53:51.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_11/index_5448_seed_24383_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-27 23:55:13.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:55:13.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:22.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_11/index_5448_seed_24384_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:56:44.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 00:07:32.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:38.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_11/index_9995_seed_21405_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 00:08:54.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:54.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:08.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_11/index_9995_seed_21406_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:10:31.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 00:21:21.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:21.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_11/index_7628_seed_29636_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 00:22:42.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:42.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:23:51.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_11/index_7628_seed_29637_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 00:24:11.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:24:12.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 00:35:07.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:13.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_11/index_4946_seed_10304_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 00:36:33.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:33.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:37:45.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_11/index_4946_seed_10305_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:38:10.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 00:49:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:08.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_11/index_4990_seed_16194_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 00:50:27.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:27.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:36.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_11/index_4990_seed_16195_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:51:57.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 01:02:48.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:54.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_11/index_7913_seed_7558_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 01:04:09.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:04:09.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:21.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_11/index_7913_seed_7559_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:05:44.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.806 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 01:16:33.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:39.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_11/index_3887_seed_31_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 01:17:59.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:59.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:11.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_11/index_3887_seed_32_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:19:31.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.637 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.554 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 01:30:58.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_11/index_6009_seed_150_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 01:32:25.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:25.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:29.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_11/index_6009_seed_151_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:33:52.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 01:44:44.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:50.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_11/index_3818_seed_32136_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 01:46:03.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:46:03.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:16.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_11/index_3818_seed_32137_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:47:37.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 01:58:27.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:33.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_11/index_4549_seed_29501_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 01:59:45.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:45.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:00:58.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_11/index_4549_seed_29502_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 02:01:17.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:01:18.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 02:12:08.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:14.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_11/index_2038_seed_12404_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 02:13:31.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:31.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:14:41.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_11/index_2038_seed_12405_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:15:04.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 02:25:53.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_11/index_6065_seed_2690_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 02:27:12.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:27:12.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:25.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_11/index_6065_seed_2691_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:28:45.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 02:39:38.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:44.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_11/index_3881_seed_4576_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 02:41:00.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:41:00.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:08.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_11/index_3881_seed_4577_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:42:31.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 02:53:20.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_11/index_3768_seed_8273_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 02:54:43.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:43.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:55:51.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_11/index_3768_seed_8274_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:56:13.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-28 02:58:07.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 03:07:08.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_11/index_10_seed_2694_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 03:08:28.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:28.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:09:38.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_11/index_10_seed_2695_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 03:10:01.813 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:10:03.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 03:20:53.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:59.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_11/index_4864_seed_15451_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 03:22:14.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:22:14.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:26.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_11/index_4864_seed_15452_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:23:49.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 03:34:40.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_11/index_350_seed_95_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 03:36:06.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:36:06.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:12.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_11/index_350_seed_96_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:37:35.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.958 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 03:50:13.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_11/index_4448_seed_17314_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 03:51:34.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:34.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:52:46.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_11/index_4448_seed_17315_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:53:06.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 04:04:03.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_11/index_2213_seed_13042_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 04:05:22.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:22.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:33.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_11/index_2213_seed_13043_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:06:56.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 04:17:48.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_11/index_6287_seed_14854_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 04:19:09.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:19:09.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:20.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_11/index_6287_seed_14855_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:20:47.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.283 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 04:31:40.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:46.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_11/index_8428_seed_28340_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 04:33:01.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:33:01.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:09.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_11/index_8428_seed_28341_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:34:29.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 04:45:20.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:26.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_11/index_9446_seed_16262_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 04:46:42.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:42.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:47:50.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_11/index_9446_seed_16263_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:48:11.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 04:59:03.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:09.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_11/index_2098_seed_16752_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 05:00:26.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:26.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:32.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_11/index_2098_seed_16753_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:01:52.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 05:12:44.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_11/index_9604_seed_14249_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 05:14:06.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:14:06.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:15.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_11/index_9604_seed_14250_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:15:35.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 05:26:25.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_11/index_4088_seed_24153_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 05:27:47.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:47.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:28:56.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_11/index_4088_seed_24154_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:29:22.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 05:40:15.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_11/index_715_seed_16737_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 05:41:36.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:36.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:42:47.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_11/index_715_seed_16738_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 05:43:09.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:43:10.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 05:54:06.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_11/index_3864_seed_483_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 05:55:28.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:28.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_11/index_3864_seed_484_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:57:00.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.682 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 06:08:27.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:33.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_11/index_1926_seed_1223_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 06:09:52.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:52.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:10:58.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_11/index_1926_seed_1224_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:11:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 06:22:11.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_11/index_4751_seed_28410_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 06:23:34.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:34.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:24:42.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_11/index_4751_seed_28411_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 06:25:06.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:25:07.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 06:36:02.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:08.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_11/index_6918_seed_27077_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 06:37:21.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:21.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:32.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_11/index_6918_seed_27078_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:38:53.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 06:49:45.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:51.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_11/index_2582_seed_24925_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 06:51:06.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:51:06.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:16.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_11/index_2582_seed_24926_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:52:37.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 07:03:27.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:33.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_11/index_1942_seed_4148_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 07:04:54.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:54.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:05:59.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_11/index_1942_seed_4149_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:06:21.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 07:17:12.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_11/index_5244_seed_13239_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 07:18:34.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:34.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:19:42.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_11/index_5244_seed_13240_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:20:04.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 07:30:53.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:59.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_11/index_7816_seed_17135_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 07:32:18.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:32:18.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:26.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_11/index_7816_seed_17136_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:33:48.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 07:44:40.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:46.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_11/index_3008_seed_22803_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 07:46:03.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:46:03.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:12.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_11/index_3008_seed_22804_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 07:47:32.045 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:47:33.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 07:58:21.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_11/index_421_seed_10724_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 07:59:44.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:44.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:00:52.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_11/index_421_seed_10725_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:01:16.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 08:12:08.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:14.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_11/index_5090_seed_8476_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 08:13:30.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:30.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:14:38.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_11/index_5090_seed_8477_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 08:15:02.254 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:15:03.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.379 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 08:26:31.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:37.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_11/index_6144_seed_31408_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 08:27:50.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:50.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:02.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_11/index_6144_seed_31409_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:29:27.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 08:40:22.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:23.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:29.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_11/index_6992_seed_11758_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 08:41:43.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:43.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:42:53.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_11/index_6992_seed_11759_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:43:14.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 08:54:08.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_11/index_9648_seed_20714_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 08:55:29.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:29.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:38.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_11/index_9648_seed_20715_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:57:00.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 09:07:50.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_11/index_4858_seed_11716_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 09:09:12.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:09:12.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:21.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_11/index_4858_seed_11717_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:10:44.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 09:21:32.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:38.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_11/index_6704_seed_31116_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 09:22:59.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:59.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:04.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_11/index_6704_seed_31117_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:24:28.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 09:36:35.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_11/index_4853_seed_3110_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 09:37:55.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:55.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:05.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_11/index_4853_seed_3111_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:39:27.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 09:50:22.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:28.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_11/index_3089_seed_15016_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 09:51:43.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:43.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:52:55.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_11/index_3089_seed_15017_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:53:25.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 10:04:16.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:22.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_11/index_5786_seed_18455_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 10:05:35.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:35.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:06:50.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_11/index_5786_seed_18456_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:07:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 10:18:09.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:15.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_11/index_9997_seed_3571_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 10:19:29.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:29.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:20:40.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_11/index_9997_seed_3572_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:21:02.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 10:31:54.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_11/index_4091_seed_1453_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 10:33:15.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:33:15.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:26.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_11/index_4091_seed_1454_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:34:54.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:19.075 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 10:46:20.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_11/index_4897_seed_3411_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 10:47:43.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:43.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:48:51.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_11/index_4897_seed_3412_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:49:13.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 11:00:08.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_11/index_7618_seed_27191_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 11:01:31.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:31.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:39.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_11/index_7618_seed_27192_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:03:00.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 11:13:51.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:57.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_11/index_358_seed_4506_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 11:15:12.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:15:12.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:21.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_11/index_358_seed_4507_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 11:16:42.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:16:43.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 11:27:32.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:38.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_11/index_321_seed_18007_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 11:28:53.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:53.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:02.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_11/index_321_seed_18008_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:30:23.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 11:41:15.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:21.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_11/index_6778_seed_18768_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 11:42:37.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:37.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:43:47.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_11/index_6778_seed_18769_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:44:08.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 11:55:03.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:09.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_11/index_2072_seed_9796_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 11:56:28.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:28.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:34.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_11/index_2072_seed_9797_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:57:57.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 12:08:48.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:54.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_11/index_9045_seed_5375_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 12:10:09.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:10:09.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:25.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_11/index_9045_seed_5376_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:11:48.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 12:22:40.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_11/index_398_seed_8057_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 12:24:01.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:24:01.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:13.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_11/index_398_seed_8058_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:25:34.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 12:36:24.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:30.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_11/index_6611_seed_14841_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 12:37:47.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:47.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:38:55.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_11/index_6611_seed_14842_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:39:17.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 12:50:08.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:09.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_11/index_3538_seed_976_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 12:51:28.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:28.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:52:41.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_11/index_3538_seed_977_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:53:04.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:31.102 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 13:04:33.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:40.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_11/index_3916_seed_8936_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 13:06:03.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:06:03.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:07.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_11/index_3916_seed_8937_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:07:34.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 13:18:22.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_11/index_6906_seed_20345_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 13:19:44.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:44.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:20:53.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_11/index_6906_seed_20346_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:21:14.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 13:32:07.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_11/index_9198_seed_17842_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 13:33:31.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:31.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:38.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_11/index_9198_seed_17843_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:35:00.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 13:45:49.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_11/index_4970_seed_4686_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 13:47:09.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:47:09.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:22.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_11/index_4970_seed_4687_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:48:42.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 13:59:30.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_11/index_1320_seed_21356_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 14:00:54.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:54.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:00.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_11/index_1320_seed_21357_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:02:21.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 14:13:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:14.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_11/index_22_seed_595_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 14:14:29.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:29.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:15:42.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_11/index_22_seed_596_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:16:09.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 14:27:02.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_11/index_8391_seed_17553_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 14:28:24.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:24.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:31.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_11/index_8391_seed_17554_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:29:51.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 14:40:44.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_11/index_8911_seed_20857_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 14:42:03.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:42:03.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:15.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_11/index_8911_seed_20858_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:43:36.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 14:54:25.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:31.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_11/index_9156_seed_18395_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 14:55:45.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:45.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:56:55.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_11/index_9156_seed_18396_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:57:15.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 15:08:07.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_11/index_5782_seed_6188_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 15:09:29.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:29.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:10:39.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_11/index_5782_seed_6189_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:11:02.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.496 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 15:23:36.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:42.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_11/index_3049_seed_16640_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 15:24:57.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:57.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:08.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_11/index_3049_seed_16641_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:26:32.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 15:37:21.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_11/index_2343_seed_5690_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 15:38:44.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:44.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:39:52.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_11/index_2343_seed_5691_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:40:14.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 15:51:08.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:14.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_11/index_3769_seed_17461_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 15:52:28.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:28.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:53:40.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_11/index_3769_seed_17462_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:54:04.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 16:04:54.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:06:00.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_11/index_7349_seed_32454_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 16:06:17.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:06:17.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_11/index_7349_seed_32455_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:07:50.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:40.002 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 16:18:40.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_11/index_6870_seed_21902_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 16:20:00.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:20:00.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:12.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_11/index_6870_seed_21903_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:21:34.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 16:32:23.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:29.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_11/index_9981_seed_16884_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 16:33:47.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:47.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:34:53.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_11/index_9981_seed_16885_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 16:35:13.431 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:35:14.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 16:46:07.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:13.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_11/index_9634_seed_9492_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 16:47:28.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:28.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:39.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_11/index_9634_seed_9493_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 16:48:59.323 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:49:00.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 16:59:50.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_11/index_8373_seed_11077_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 17:01:10.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:01:10.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:21.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_11/index_8373_seed_11078_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:02:44.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 17:13:34.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_11/index_9969_seed_3583_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 17:14:55.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:55.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:08.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_11/index_9969_seed_3584_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:16:32.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 17:27:21.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_11/index_7566_seed_565_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 17:28:40.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:40.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:29:53.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_11/index_7566_seed_566_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:30:18.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.955 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 17:41:44.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_11/index_1778_seed_32951_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 17:43:06.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:43:06.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:21.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_11/index_1778_seed_32952_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:44:43.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 17:55:31.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:32.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:37.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_11/index_7343_seed_5278_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 17:56:53.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:53.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:02.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_11/index_7343_seed_5279_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:58:25.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 18:09:15.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:15.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_11/index_223_seed_13110_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 18:10:36.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:36.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:11:49.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_11/index_223_seed_13111_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:12:10.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 18:23:03.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:09.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_11/index_2272_seed_24432_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 18:24:26.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:26.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:33.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_11/index_2272_seed_24433_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:25:54.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 18:36:44.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_11/index_9988_seed_7819_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 18:38:06.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:38:06.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:16.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_11/index_9988_seed_7820_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:39:39.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 18:50:26.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:32.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_11/index_898_seed_10383_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 18:51:46.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:46.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:52:59.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_11/index_898_seed_10384_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:53:22.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 19:04:14.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:14.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_11/index_5267_seed_32023_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 19:05:33.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:33.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:06:45.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_11/index_5267_seed_32024_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:07:08.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 19:18:02.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_11/index_9049_seed_11220_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 19:19:23.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:23.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:30.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_11/index_9049_seed_11221_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:20:53.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 19:31:44.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:50.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_11/index_1889_seed_9263_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 19:33:06.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:33:06.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:16.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_11/index_1889_seed_9264_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:34:37.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 19:45:24.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:24.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:31.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_11/index_9589_seed_12727_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 19:46:47.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:47.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:47:56.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_11/index_9589_seed_12728_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:48:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.633 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 19:59:44.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:50.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_11/index_146_seed_29850_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 20:01:04.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:01:04.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:15.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_11/index_146_seed_29851_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 20:02:34.357 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:02:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 20:13:24.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_11/index_5252_seed_11177_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 20:14:47.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:47.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:15:55.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_11/index_5252_seed_11178_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 20:16:15.048 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:16:16.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 20:27:07.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_11/index_9274_seed_1874_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 20:28:27.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:27.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:38.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_11/index_9274_seed_1875_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:29:59.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 20:40:49.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_11/index_4155_seed_10494_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 20:42:09.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:42:09.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:22.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_11/index_4155_seed_10495_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:43:46.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 20:54:36.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:42.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_11/index_2518_seed_14907_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 20:55:59.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:59.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:09.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_11/index_2518_seed_14908_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:57:34.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 21:09:41.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_11/index_8727_seed_7305_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 21:11:04.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:11:04.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:14.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_11/index_8727_seed_7306_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:12:35.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 21:23:24.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:30.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_11/index_4043_seed_16113_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 21:24:42.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:42.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:25:53.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_11/index_4043_seed_16114_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:26:15.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 21:37:07.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:13.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_11/index_8643_seed_32285_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 21:38:30.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:30.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:36.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_11/index_8643_seed_32286_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:39:58.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:48.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 21:50:48.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_11/index_2086_seed_14157_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 21:52:10.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:52:10.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:21.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_11/index_2086_seed_14158_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 21:53:42.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:53:43.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 22:04:32.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_11/index_3657_seed_10982_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 22:05:53.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:53.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:04.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_11/index_3657_seed_10983_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:07:26.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.742 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 22:18:53.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:59.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_11/index_8154_seed_18497_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 22:20:18.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:20:18.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:24.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_11/index_8154_seed_18498_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:21:47.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 22:32:40.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:46.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_11/index_5396_seed_21964_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 22:34:04.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:34:04.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:12.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_11/index_5396_seed_21965_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:35:35.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 22:46:24.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:30.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_11/index_2810_seed_30739_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 22:47:44.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:44.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:48:56.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_11/index_2810_seed_30740_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:49:18.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 23:00:07.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_11/index_6264_seed_1813_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 23:01:31.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:31.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:38.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_11/index_6264_seed_1814_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:02:59.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 23:13:49.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:50.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:55.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_11/index_606_seed_19485_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 23:15:09.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:15:09.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:20.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_11/index_606_seed_19486_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:16:44.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 23:27:32.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:38.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_11/index_89_seed_23058_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 23:28:52.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:52.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:02.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_11/index_89_seed_23059_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 23:30:22.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:30:23.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 23:41:14.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_11/index_2848_seed_5483_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 23:42:36.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:36.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:43:44.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_11/index_2848_seed_5484_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:44:06.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-28 23:55:00.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_11/index_168_seed_7000_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-28 23:56:19.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:19.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:29.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_11/index_168_seed_7001_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:57:49.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.806 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.656 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-29 00:08:39.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:45.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_11/index_9314_seed_13247_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-29 00:10:00.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:10:00.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:08.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_11/index_9314_seed_13248_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:11:29.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-29 00:22:19.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:25.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_11/index_3381_seed_22010_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-29 00:23:40.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:40.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:24:52.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_11/index_3381_seed_22011_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:25:13.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.623 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-29 00:36:43.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:49.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_11/index_469_seed_28688_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-29 00:38:02.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:38:02.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:15.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_11/index_469_seed_28689_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:39:39.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-29 00:50:27.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:33.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_11/index_3826_seed_1555_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-29 00:51:46.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:46.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:52:56.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_11/index_3826_seed_1556_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:53:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 46, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 46, dp_rank 23: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=11, rank_in_group=1 +2026-06-29 01:04:09.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:15.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_11/index_4670_seed_1250_dp_23_rank_46_sample_2_slot_0.mp4 +2026-06-29 01:05:30.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 46 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:30.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 46 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:06:41.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 46: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_11/index_4670_seed_1251_dp_23_rank_46_sample_3_slot_1.mp4 +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 46: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 01:07:04.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 46: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 46: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 46, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_48.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_48.log new file mode 100644 index 0000000000000000000000000000000000000000..45f20e651334e94185daf8fcce0217ba3fa4b2e5 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_48.log @@ -0,0 +1,6542 @@ +2026-05-29 20:45:41.484 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-29 20:45:41.486 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-29 20:45:42.166 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:45:42.167 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 20:51:46.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26130_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 20:53:28.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:28.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:38.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26131_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 20:55:07.087 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 20:55:17.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 48, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 21:06:57.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:04.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26455_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 21:08:17.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:17.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:09:39.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26456_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:10:01.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 21:10:01.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 21:21:17.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1344_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 21:22:38.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:38.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:23:53.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1345_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:24:15.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 21:24:15.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 21:35:35.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1782_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 21:36:57.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:57.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:14.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1783_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:38:35.729 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 21:38:36.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 21:49:56.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26284_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 21:51:18.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:18.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:52:35.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26285_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:53:00.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 21:53:01.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 22:04:18.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28027_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 22:05:40.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:40.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:06:52.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28028_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:07:14.436 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 22:07:15.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 22:18:35.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:42.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19486_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 22:19:55.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:55.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:14.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19487_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:21:37.450 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 22:21:38.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 22:32:58.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:58.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16829_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 22:34:24.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:24.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:36.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16830_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:36:00.133 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 22:36:01.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 22:47:20.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6180_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 22:48:42.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:42.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:49:56.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6181_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:50:19.421 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 22:50:20.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 23:01:44.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29911_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 23:03:06.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:03:06.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:23.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29912_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:04:48.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 23:04:49.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:53.431 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 23:16:57.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27365_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 23:18:22.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:22.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:35.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27366_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:19:57.201 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 23:19:57.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 23:31:14.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:21.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8789_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 23:32:33.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:33.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:33:53.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8790_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:34:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 23:34:15.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 23:45:29.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7434_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-29 23:46:50.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:50.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:09.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7435_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:48:30.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-29 23:48:31.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-29 23:59:50.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29111_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 00:01:13.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:01:13.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:26.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29112_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:02:54.393 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 00:02:55.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 00:14:13.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:20.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1985_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 00:15:44.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:44.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:16:51.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1986_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:17:17.280 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 00:17:18.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 00:28:36.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32330_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 00:30:01.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:30:01.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:16.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32331_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:31:38.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 00:31:39.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 00:42:58.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20819_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 00:44:21.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:21.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:45:38.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20820_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:46:01.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 00:46:02.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 00:57:15.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9953_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 00:58:37.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:37.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:59:51.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9954_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:00:15.814 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 01:00:16.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 01:11:38.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:46.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6671_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 01:12:59.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:59.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:18.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6672_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:14:43.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 01:14:44.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 01:26:05.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:14.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7231_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 01:27:31.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:31.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:28:42.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7232_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:29:03.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 01:29:04.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.888 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 01:41:19.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18721_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 01:42:44.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:44.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:43:56.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18722_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:44:17.853 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 01:44:18.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 01:55:38.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:45.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9948_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 01:56:59.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:59.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:14.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9949_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:58:38.979 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 01:58:39.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 02:09:58.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21357_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 02:11:17.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:17.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:12:39.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21358_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:13:02.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 02:13:02.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 02:24:23.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16558_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 02:25:44.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:44.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:05.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16559_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:27:32.759 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 02:27:33.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.75s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 02:38:56.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_25_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 02:40:21.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:21.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:35.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_26_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:41:57.174 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 02:41:57.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 02:54:31.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15322_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 02:55:55.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:55.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:08.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15323_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:57:30.368 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 02:57:31.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.342 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 03:08:50.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9661_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 03:10:11.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:10:11.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:27.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9662_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:11:52.681 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 03:11:53.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 03:23:09.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:17.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20497_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 03:24:34.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:34.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:25:46.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20498_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 03:26:07.442 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:26:07.593 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 03:26:08.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 03:37:26.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14885_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 03:38:52.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:52.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:07.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14886_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:40:38.392 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 03:40:39.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 03:51:59.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28203_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 03:53:24.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:24.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:54:38.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28204_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:55:03.582 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 03:55:04.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.279 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 04:07:19.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:28.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17146_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 04:08:48.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:48.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:01.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17147_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:10:27.395 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 04:10:28.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 04:21:48.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15657_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 04:23:12.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:23:12.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:29.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15658_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:24:53.758 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 04:24:54.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 04:36:12.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10427_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 04:37:35.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:35.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:38:50.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10428_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:39:15.496 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 04:39:16.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 04:50:41.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:48.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_1997_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 04:52:01.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:52:01.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:19.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_1998_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:53:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 04:53:44.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 05:05:07.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3066_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 05:06:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:28.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:07:43.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3067_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:08:12.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 05:08:13.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 05:19:25.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:32.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12880_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 05:20:47.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:47.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:21:58.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12881_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:22:19.952 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 05:22:20.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 05:33:38.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:45.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12609_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 05:35:03.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:35:03.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:15.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12610_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:36:36.912 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 05:36:37.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 05:47:59.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8701_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 05:49:19.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:19.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:35.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8702_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:50:57.699 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 05:50:58.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 06:02:21.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:30.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2225_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 06:03:42.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:42.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:04.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2226_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:05:32.252 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 06:05:33.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 06:16:54.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:03.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6625_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 06:18:18.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:18.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:31.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6626_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 06:19:58.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:19:58.969 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 06:19:59.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:12.230 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 06:32:17.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:26.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17610_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 06:33:38.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:38.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:34:55.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17611_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:35:15.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 06:35:16.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 06:46:37.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:44.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21420_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 06:48:03.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:48:03.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:15.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21421_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:49:37.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 06:49:38.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.35s (2.01min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 07:00:59.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26055_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 07:02:22.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:22.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:38.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26056_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:04:00.113 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 07:04:00.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 07:15:21.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:22.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:29.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15758_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 07:16:46.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:46.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:06.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15759_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:18:41.089 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 07:18:41.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 07:30:08.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18243_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 07:31:31.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:31.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:32:48.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18244_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:33:10.747 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 07:33:11.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 07:44:30.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25137_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 07:45:53.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:53.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:04.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25138_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:47:29.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 07:47:30.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 07:59:00.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8942_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 08:00:24.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:24.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:01:41.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8943_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:02:08.027 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 08:02:08.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 08:13:34.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21474_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 08:14:58.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:58.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:17.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21475_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:16:43.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 08:16:44.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 08:28:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_95_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 08:29:39.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:39.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:30:55.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_96_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:31:22.834 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 08:31:23.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 08:42:53.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7486_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 08:44:17.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:17.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:45:35.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7487_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:46:00.669 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 08:46:01.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.129 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:24.800 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 08:59:29.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:40.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5949_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 09:00:55.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:55.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:11.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5950_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:02:32.185 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 09:02:33.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 09:14:04.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:12.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18296_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 09:15:27.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:27.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:16:52.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18297_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:17:16.114 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 09:17:17.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 09:28:42.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18666_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 09:30:07.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:30:07.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:27.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18667_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:31:53.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 09:31:54.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 09:43:17.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17708_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 09:44:40.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:40.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:02.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17709_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:46:24.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 09:46:25.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 09:57:46.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:54.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25590_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 09:59:11.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:59:11.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25591_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:00:51.403 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 10:00:52.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 10:12:07.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:16.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14038_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 10:13:29.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:29.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:14:44.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14039_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:15:05.987 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 10:15:06.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 10:26:24.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6818_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 10:27:47.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:47.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:02.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6819_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:29:28.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 10:29:29.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.981 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 10:40:46.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14360_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 10:42:13.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:42:13.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:28.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14361_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:43:54.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 10:43:54.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 10:55:08.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11083_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 10:56:32.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:32.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:57:52.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11084_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:58:22.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 10:58:23.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 11:09:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9322_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 11:11:07.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:11:07.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:20.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9323_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:12:42.357 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 11:12:43.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:03.143 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 11:25:08.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24001_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 11:26:30.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:30.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:27:52.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24002_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 11:28:20.076 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:28:20.188 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 11:28:20.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 11:39:42.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7436_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 11:41:04.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:41:04.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:19.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7437_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:42:41.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 11:42:42.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 11:54:06.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:14.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_958_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 11:55:30.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:30.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:56:42.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_959_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:57:05.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 11:57:06.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 12:08:30.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11095_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 12:09:54.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:54.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:07.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11096_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:11:29.136 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 12:11:30.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 12:22:53.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13337_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 12:24:17.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:17.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:30.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13338_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:25:54.712 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 12:25:55.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 12:37:12.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24632_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 12:38:38.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:38.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:39:47.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24633_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 12:40:10.052 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:40:10.139 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 12:40:11.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 12:51:30.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31473_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 12:52:57.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:57.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:11.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31474_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:54:31.853 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 12:54:32.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 13:05:59.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17024_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 13:07:21.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:21.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:08:38.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17025_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:09:03.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 13:09:04.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 13:20:24.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:32.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6379_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 13:21:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:44.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6380_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:23:29.139 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 13:23:30.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 13:34:48.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33214_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 13:36:10.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:36:10.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:25.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33215_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:37:47.920 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 13:37:48.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.616 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 13:50:02.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16433_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 13:51:25.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:25.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:52:48.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16434_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:53:11.024 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 13:53:11.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 14:04:33.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:44.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14140_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 14:05:58.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:58.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:19.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14141_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:07:58.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 14:07:59.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 14:19:43.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1066_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 14:21:15.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:21:15.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:22:39.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1067_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:23:12.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 14:23:13.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 14:34:38.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:46.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16675_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 14:36:00.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:36:00.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:16.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16676_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:37:53.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 14:37:54.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 14:49:20.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:28.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3081_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 14:50:41.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:41.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:00.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3082_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:52:25.959 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 14:52:26.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 15:05:01.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:10.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14047_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 15:06:24.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:24.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:07:40.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14048_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:08:03.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 15:08:04.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 15:19:35.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:43.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10069_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 15:20:56.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:56.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:14.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10070_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:22:37.379 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 15:22:38.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 15:34:11.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21019_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 15:35:41.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:41.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:36:53.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21020_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:37:18.406 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 15:37:19.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 15:48:48.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8836_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 15:50:11.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:50:11.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:25.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8837_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:51:49.951 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 15:51:51.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 16:03:17.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14208_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 16:04:40.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:40.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:05:57.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14209_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:06:19.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 16:06:20.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.556 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 16:18:48.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32868_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 16:20:21.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:21.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:21:33.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32869_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:22:04.793 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 16:22:05.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 16:33:33.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:43.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7822_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 16:35:01.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:35:01.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:16.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7823_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:36:39.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 16:36:40.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 16:48:15.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:24.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5777_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 16:49:40.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:40.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:50:53.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5778_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:51:15.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 16:51:16.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 17:02:55.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11196_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 17:04:21.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:21.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:33.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11197_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:05:56.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 17:05:57.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 17:17:35.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20778_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 17:19:03.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:19:03.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:15.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20779_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:20:37.483 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 17:20:38.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 17:32:10.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29857_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 17:33:33.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:33.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:00.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29858_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:35:25.790 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 17:35:26.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 17:46:54.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20791_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 17:48:17.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:17.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:49:38.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20792_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:50:01.447 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 17:50:02.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 18:01:35.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:44.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12958_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 18:02:57.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:57.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:17.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12959_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:04:39.838 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 18:04:40.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 18:16:18.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32886_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 18:17:44.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:44.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:00.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32887_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:19:31.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 18:19:32.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 18:31:18.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:19.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:30.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9541_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 18:32:44.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:44.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:14.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9542_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 18:34:45.308 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:34:45.421 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 18:34:46.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:18.493 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 18:47:24.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11840_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 18:48:49.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:49.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:12.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11841_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 18:50:38.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:50:38.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 18:50:39.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 19:02:13.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19511_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 19:03:40.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:40.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:01.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19512_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:05:34.003 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 19:05:34.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 19:17:12.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:22.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29113_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 19:18:38.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:38.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:19:59.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29114_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:20:25.883 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 19:20:26.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:53.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 19:31:55.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:06.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18913_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 19:33:21.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:21.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:34:55.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18914_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:35:23.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 19:35:24.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 19:47:03.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24556_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 19:48:26.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:26.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:49:48.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24557_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:50:10.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 19:50:11.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 20:01:46.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:57.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12255_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 20:03:12.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:03:12.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:29.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12256_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:04:55.512 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 20:04:56.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 20:16:29.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:41.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18519_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 20:17:56.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:56.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:19.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18520_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:19:46.546 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 20:19:47.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 20:31:20.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:29.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15431_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 20:32:42.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:42.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:00.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15432_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:34:23.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 20:34:24.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 20:46:03.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16699_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 20:47:26.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:26.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:48:49.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16700_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:49:14.641 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 20:49:15.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 21:00:47.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25195_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 21:02:14.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:02:14.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:31.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25196_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:03:57.572 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 21:03:58.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.034 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 21:17:31.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:41.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10949_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 21:18:53.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:53.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:13.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10950_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:20:37.120 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 21:20:38.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 21:32:09.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:49:21.312 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-30 21:49:21.312 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-30 21:49:22.005 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:49:22.006 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 21:55:22.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:49.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26130_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 21:57:08.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:57:08.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:18.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26131_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:58:57.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 21:59:08.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 48, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 22:10:52.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:12:01.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26455_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 22:12:14.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:14.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:13:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26456_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:14:02.543 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 22:14:03.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 22:25:33.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:42.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1344_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 22:26:56.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:56.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:14.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1345_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:28:36.488 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 22:28:37.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 22:40:03.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:04.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1782_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 22:41:30.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:30.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:42:45.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1783_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:43:09.986 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 22:43:10.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 22:54:38.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:38.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26284_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 22:56:01.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:56:01.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:25.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26285_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:57:49.184 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 22:57:50.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 23:09:20.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28027_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 23:10:43.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:43.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:11.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28028_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:12:35.329 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 23:12:36.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 23:24:05.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:14.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19486_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 23:25:27.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:27.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:26:47.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19487_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:27:14.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 23:27:14.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 23:38:46.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16829_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 23:40:13.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:40:13.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:25.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16830_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:41:47.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 23:41:48.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-30 23:53:10.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6180_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-30 23:54:31.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:31.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:55:52.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6181_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:56:15.011 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-30 23:56:15.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 00:07:41.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29911_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 00:09:09.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:09:09.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:24.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29912_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:10:45.433 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 00:10:46.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:54.224 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 00:22:59.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27365_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 00:24:25.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:25.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:25:40.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27366_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:26:04.448 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 00:26:05.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 00:37:30.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8789_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 00:38:53.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:53.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:18.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8790_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:40:45.482 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 00:40:46.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 00:52:07.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7434_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 00:53:28.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:28.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:54:49.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7435_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:55:21.225 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 00:55:22.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 01:06:44.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29111_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 01:08:07.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:08:07.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:26.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29112_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:09:51.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 01:09:52.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 01:21:16.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1985_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 01:22:43.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:43.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:23:55.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1986_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:24:19.475 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 01:24:20.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 01:35:36.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32330_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 01:36:58.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:58.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:15.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32331_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:38:37.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 01:38:38.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 01:50:03.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20819_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 01:51:24.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:24.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:52:42.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20820_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 01:53:05.979 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:53:06.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 01:53:07.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 02:04:34.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9953_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 02:06:01.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:06:01.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:13.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9954_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:07:38.911 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 02:07:39.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 48, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-05-31 02:19:01.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6671_dp_24_rank_48_sample_0_slot_0.mp4 +2026-05-31 02:20:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:22.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:21:38.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6672_dp_24_rank_48_sample_1_slot_1.mp4 +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:22:01.806 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-05-31 02:22:02.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:03:35.608 | INFO | __main__:__init__:946 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-06-25 18:03:35.609 | INFO | __main__:__init__:947 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-06-25 18:03:36.284 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:03:36.285 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 18:04:09.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26130_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 18:05:49.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:49.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:06:56.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26131_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:07:22.226 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 18:07:32.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 48, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 18:18:41.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:47.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26455_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 18:20:01.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:20:01.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:12.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26456_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:21:34.304 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 18:21:35.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 18:32:26.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1344_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 18:33:45.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:45.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:34:54.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1345_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:35:14.705 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 18:35:15.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 18:46:08.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1782_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 18:47:27.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:27.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:48:38.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1783_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:49:03.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 18:49:03.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 19:00:02.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26284_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 19:01:23.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:23.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:34.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26285_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:02:56.806 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 19:02:57.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 19:13:49.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28027_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 19:15:10.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:15:10.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:20.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28028_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:16:41.774 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 19:16:42.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 19:27:32.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19486_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 19:28:50.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:50.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:03.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19487_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:30:23.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 19:30:24.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 19:41:17.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16829_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 19:42:39.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:39.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:43:50.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16830_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:44:14.791 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 19:44:15.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 19:55:08.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6180_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 19:56:28.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:57:41.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6181_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:58:02.250 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 19:58:03.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 20:09:02.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29911_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 20:10:23.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:23.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:35.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29912_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 20:11:59.727 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:11:59.861 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 20:12:00.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.905 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 20:23:32.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27365_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 20:24:54.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:54.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:05.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27366_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:26:25.483 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 20:26:26.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 20:37:18.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8789_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 20:38:38.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:38.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:39:48.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8790_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:40:08.645 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 20:40:09.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 20:51:03.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7434_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 20:52:23.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:23.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:35.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7435_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:53:56.300 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 20:53:57.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 21:04:50.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:51.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29111_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 21:06:13.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:06:13.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:23.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29112_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:07:48.837 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 21:07:49.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 21:18:43.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:44.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:49.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1985_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 21:20:09.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:20:09.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:17.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1986_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:21:38.899 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 21:21:39.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 21:32:31.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32330_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 21:33:50.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:50.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:01.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32331_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:35:20.766 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 21:35:21.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 21:46:16.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:22.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20819_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 21:47:37.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:37.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:48:47.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20820_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:49:09.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 21:49:10.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 22:00:05.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9953_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 22:01:25.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:25.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:37.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9954_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:02:57.906 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 22:02:58.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.002 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 22:13:50.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6671_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 22:15:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:15:10.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:22.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6672_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:16:44.839 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 22:16:45.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 22:27:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7231_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 22:29:05.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:29:05.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:10.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7232_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:30:31.234 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 22:30:32.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:58.205 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 22:42:00.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:07.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18721_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 22:43:22.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:22.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:31.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18722_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:44:51.497 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 22:44:52.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 22:55:45.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:51.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9948_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 22:57:05.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:57:05.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:16.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9949_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:58:38.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 22:58:39.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 23:09:30.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21357_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 23:10:51.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:51.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:01.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21358_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:12:21.745 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 23:12:22.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.283 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 23:23:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16558_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 23:24:35.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:35.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:25:47.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16559_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:26:10.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 23:26:11.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 23:37:07.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_25_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 23:38:30.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:30.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:39:38.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_26_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:40:01.128 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 23:40:02.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-25 23:51:52.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:58.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15322_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-25 23:53:18.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:53:18.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:23.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15323_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:54:43.082 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-25 23:54:43.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 00:05:36.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:42.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9661_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 00:06:56.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:56.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:07.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9662_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:08:27.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 00:08:28.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 00:19:21.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20497_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 00:20:44.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:44.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:21:50.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20498_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:22:10.538 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 00:22:11.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 00:33:07.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:13.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14885_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 00:34:31.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:31.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:35:37.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14886_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:36:03.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 00:36:04.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 00:47:03.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:09.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28203_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 00:48:33.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:33.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:37.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28204_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:50:00.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 00:50:00.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:29.370 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 01:01:30.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:38.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17146_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 01:02:56.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:56.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:04.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17147_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:04:24.849 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 01:04:25.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 01:15:17.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15657_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 01:16:38.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:38.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:17:50.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15658_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:18:11.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 01:18:11.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 01:29:08.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:14.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10427_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 01:30:28.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:28.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:31:40.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10428_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:32:02.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 01:32:03.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 01:42:58.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_1997_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 01:44:16.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:16.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:29.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_1998_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:45:49.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 01:45:50.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 01:56:44.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:50.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3066_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 01:58:07.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:58:07.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:18.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3067_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:59:39.329 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 01:59:40.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 02:10:30.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12880_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 02:11:52.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:52.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:03.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12881_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:13:23.895 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 02:13:25.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 02:24:16.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12609_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 02:25:40.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:40.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:26:46.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12610_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:27:11.028 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 02:27:11.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 02:38:08.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:14.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8701_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 02:39:29.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:29.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:37.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8702_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 02:40:59.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:40:59.276 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 02:41:00.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 02:51:53.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2225_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 02:53:16.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:53:16.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:25.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2226_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:54:47.512 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 02:54:48.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 03:05:40.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:41.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6625_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 03:07:03.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:07:03.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:15.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6626_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:08:37.658 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 03:08:38.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.450 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 03:20:06.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17610_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 03:21:28.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:28.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:39.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17611_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:22:58.525 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 03:22:59.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 03:33:52.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21420_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 03:35:12.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:35:12.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:24.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21421_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:36:45.329 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 03:36:46.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 03:47:40.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26055_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 03:49:03.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:49:03.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:10.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26056_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:50:30.363 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 03:50:31.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 04:01:21.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15758_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 04:02:43.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:43.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:03:52.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15759_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:04:11.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 04:04:12.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 04:15:09.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18243_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 04:16:30.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:30.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:17:42.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18244_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:18:04.898 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 04:18:05.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 04:29:03.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:09.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25137_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 04:30:23.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:23.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:33.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25138_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:31:53.334 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 04:31:54.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 04:42:46.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8942_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 04:44:07.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:44:07.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:16.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8943_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:45:34.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 04:45:35.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 04:56:27.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21474_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 04:57:49.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:49.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:58:58.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21475_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:59:18.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 04:59:19.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 05:10:10.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_95_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 05:11:30.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:30.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:12:41.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_96_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:13:02.806 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 05:13:04.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 05:24:03.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:09.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7486_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 05:25:27.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:27.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:35.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7487_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:26:58.065 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 05:26:58.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:25.719 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 05:39:29.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5949_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 05:40:51.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:51.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:00.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5950_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:42:19.511 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 05:42:20.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 05:53:10.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:16.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18296_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 05:54:33.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:33.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:39.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18297_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:56:00.012 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 05:56:00.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 06:06:52.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:58.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18666_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 06:08:13.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:08:13.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:22.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18667_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:09:44.084 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 06:09:44.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 06:20:40.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17708_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 06:22:02.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:22:02.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:09.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17709_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:23:30.559 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 06:23:31.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 06:34:21.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25590_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 06:35:45.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:45.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:36:51.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25591_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:37:10.680 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 06:37:11.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 06:48:09.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14038_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 06:49:29.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:29.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:50:44.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14039_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:51:02.758 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 06:51:03.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 07:02:03.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6818_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 07:03:21.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:21.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:34.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6819_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:04:54.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 07:04:55.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 07:15:48.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:54.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14360_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 07:17:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:17:14.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:21.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14361_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:18:40.985 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 07:18:41.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 07:29:32.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11083_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 07:30:53.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:53.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:04.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11084_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:32:30.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 07:32:31.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 07:43:21.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:27.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9322_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 07:44:44.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:44.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:45:53.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9323_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:46:14.695 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 07:46:15.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.327 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 07:57:45.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:45.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24001_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 07:59:06.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:59:06.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:18.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24002_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:00:39.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 08:00:40.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 08:11:33.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7436_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 08:12:52.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:52.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:02.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7437_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:14:25.636 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 08:14:26.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 08:25:18.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:24.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_958_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 08:26:39.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:39.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:27:48.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_959_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:28:07.176 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 08:28:08.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 08:39:03.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11095_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 08:40:31.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:31.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:36.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11096_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:41:56.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 08:41:57.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 08:52:50.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13337_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 08:54:13.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:54:13.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:22.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13338_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:55:44.378 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 08:55:45.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.550 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 09:06:39.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:45.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24632_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 09:08:05.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:08:05.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:11.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24633_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:09:32.556 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 09:09:33.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 09:20:22.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:28.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31473_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 09:21:46.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:46.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:22:52.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31474_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:23:11.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 09:23:12.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 09:34:09.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:15.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17024_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 09:35:30.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:30.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:36:41.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17025_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:37:03.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 09:37:04.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 09:47:56.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6379_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 09:49:18.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:18.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:29.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6380_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:50:50.513 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 09:50:51.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 10:01:45.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33214_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 10:03:06.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:03:06.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:15.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33215_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:04:34.159 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 10:04:34.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.876 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 10:16:01.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:02.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16433_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 10:17:21.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:21.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:34.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16434_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:18:58.265 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 10:18:59.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 10:29:52.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:58.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14140_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 10:31:14.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:31:14.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:23.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14141_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:32:43.462 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 10:32:44.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 10:43:37.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1066_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 10:44:57.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:57.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:08.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1067_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:46:28.426 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 10:46:29.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 10:57:21.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16675_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 10:58:41.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:41.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:59:56.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16676_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:00:17.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 11:00:18.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 11:11:08.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3081_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 11:12:27.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:27.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:39.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3082_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:13:58.579 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 11:13:59.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 11:26:08.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:14.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14047_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 11:27:28.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:28.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:28:40.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14048_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:29:02.298 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 11:29:03.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 11:39:55.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10069_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 11:41:13.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:13.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:24.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10070_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:42:43.735 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 11:42:44.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 11:53:37.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21019_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 11:54:57.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:57.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:07.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21020_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:56:30.376 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 11:56:31.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 12:07:21.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8836_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 12:08:43.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:43.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:09:51.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8837_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:10:11.445 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 12:10:12.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 12:21:06.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:07.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14208_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 12:22:27.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:27.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:39.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14209_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:24:00.017 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 12:24:00.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.676 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 12:35:30.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:38.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32868_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 12:36:56.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:56.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:03.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32869_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:38:24.447 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 12:38:25.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 12:49:16.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7822_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 12:50:35.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:35.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:51:48.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7823_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:52:08.912 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 12:52:10.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 13:03:03.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5777_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 13:04:23.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:23.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:33.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5778_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:05:52.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 13:05:53.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 13:16:44.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11196_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 13:18:09.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:18:09.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:16.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11197_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:19:40.952 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 13:19:41.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 13:30:31.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20778_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 13:31:54.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:54.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:05.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20779_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:33:24.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 13:33:25.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 13:44:15.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29857_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 13:45:36.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:36.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:46:49.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29858_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:47:12.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 13:47:13.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 13:58:07.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20791_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 13:59:28.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:28.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:36.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20792_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:00:56.151 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 14:00:56.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 14:11:49.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12958_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 14:13:08.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:13:08.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:18.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12959_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:14:38.318 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 14:14:39.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 14:25:28.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32886_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 14:26:45.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:45.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:02.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32887_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:28:24.393 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 14:28:25.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 14:39:15.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9541_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 14:40:35.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:35.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:41:50.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9542_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:42:12.254 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 14:42:13.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.312 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 14:53:45.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11840_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 14:55:06.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:55:06.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:16.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11841_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:56:35.712 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 14:56:36.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 15:07:27.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19511_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 15:08:46.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:46.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:09:56.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19512_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:10:19.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 15:10:19.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 15:21:10.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29113_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 15:22:30.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:30.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:23:41.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29114_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:24:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 15:24:02.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 15:34:56.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18913_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 15:36:16.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:16.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:30.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18914_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:37:49.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 15:37:50.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 15:48:45.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24556_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 15:50:04.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:50:04.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:15.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24557_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:51:34.854 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 15:51:35.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 16:02:26.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12255_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 16:03:47.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:47.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:04:57.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12256_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:05:17.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 16:05:18.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 16:16:06.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18519_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 16:17:32.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:32.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:18:42.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18520_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:19:03.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 16:19:04.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 16:29:54.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15431_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 16:31:13.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:13.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:24.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15432_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:32:45.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 16:32:45.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 16:43:38.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:44.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16699_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 16:44:57.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:57.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:08.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16700_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:46:30.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 16:46:31.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 16:57:20.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25195_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 16:58:41.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:41.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:59:54.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25196_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:00:15.118 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 17:00:15.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:50.954 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 17:12:52.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10949_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 17:14:13.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:14:13.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:25.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10950_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:15:47.212 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 17:15:48.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 17:26:39.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:45.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_12/index_1054_seed_32581_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 17:27:59.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:59.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:09.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_12/index_1054_seed_32582_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:29:29.539 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 17:29:30.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 17:40:20.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:26.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_12/index_9232_seed_25731_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 17:41:45.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:45.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:42:54.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_12/index_9232_seed_25732_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:43:13.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 17:43:14.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 17:54:06.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_12/index_9258_seed_5214_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 17:55:25.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:25.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:38.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_12/index_9258_seed_5215_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:56:58.119 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 17:56:59.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 18:07:49.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:55.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_12/index_5642_seed_7590_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 18:09:14.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:09:14.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:21.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_12/index_5642_seed_7591_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:10:45.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 18:10:45.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 18:21:38.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_12/index_537_seed_12603_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 18:22:57.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:57.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:12.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_12/index_537_seed_12604_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:24:36.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 18:24:37.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.994 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 18:35:27.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_12/index_3311_seed_18936_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 18:36:47.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:47.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:03.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_12/index_3311_seed_18937_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:38:23.789 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 18:38:24.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 18:49:15.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_12/index_1849_seed_1097_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 18:50:34.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:34.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:51:45.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_12/index_1849_seed_1098_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:52:04.233 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 18:52:05.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 19:03:02.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_12/index_4467_seed_17951_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 19:04:26.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:26.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:35.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_12/index_4467_seed_17952_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:05:54.851 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 19:05:55.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 19:16:46.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:52.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_12/index_5836_seed_7701_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 19:18:12.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:18:12.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:20.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_12/index_5836_seed_7702_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:19:39.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 19:19:40.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.466 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 19:31:08.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_12/index_6728_seed_3946_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 19:32:32.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:32.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:33:43.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_12/index_6728_seed_3947_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:34:06.252 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 19:34:07.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 19:45:03.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_12/index_9620_seed_15359_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 19:46:22.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:22.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:32.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_12/index_9620_seed_15360_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:47:51.889 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 19:47:52.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 19:58:45.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:45.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_12/index_9972_seed_4734_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 20:00:04.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:00:04.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:17.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_12/index_9972_seed_4735_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:01:37.191 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 20:01:38.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 20:12:28.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_12/index_6474_seed_23918_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 20:13:51.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:51.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:14:59.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_12/index_6474_seed_23919_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:15:21.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 20:15:22.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 20:26:14.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:20.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_12/index_5337_seed_31078_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 20:27:33.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:33.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:28:48.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_12/index_5337_seed_31079_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:29:11.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 20:29:12.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 20:40:06.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:06.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_12/index_9961_seed_11394_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 20:41:23.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:23.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:42:39.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_12/index_9961_seed_11395_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:43:00.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 20:43:01.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 20:53:52.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_12/index_5940_seed_12275_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 20:55:17.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:55:17.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:25.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_12/index_5940_seed_12276_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:56:53.304 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 20:56:54.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 21:07:44.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:50.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_12/index_2476_seed_19988_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 21:09:04.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:09:04.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:18.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_12/index_2476_seed_19989_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:10:41.834 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 21:10:42.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 21:21:31.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_12/index_8267_seed_16723_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 21:22:55.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:55.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:02.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_12/index_8267_seed_16724_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:24:26.029 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 21:24:26.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 21:35:16.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_12/index_2563_seed_18233_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 21:36:36.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:36.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:37:49.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_12/index_2563_seed_18234_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:38:10.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 21:38:11.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.065 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 21:49:43.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_12/index_3464_seed_3230_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 21:51:04.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:51:04.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:15.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_12/index_3464_seed_3231_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:52:37.867 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 21:52:38.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 22:03:26.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:27.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_12/index_3563_seed_7369_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 22:04:46.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:46.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:05:56.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_12/index_3563_seed_7370_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:06:17.099 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 22:06:18.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 22:17:07.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:13.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_12/index_2956_seed_20295_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 22:18:27.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:27.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:37.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_12/index_2956_seed_20296_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:19:58.909 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 22:19:59.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 22:30:49.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:55.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_12/index_288_seed_17070_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 22:32:15.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:32:15.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:20.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_12/index_288_seed_17071_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:33:42.850 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 22:33:43.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 22:44:32.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_12/index_677_seed_16515_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 22:45:53.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:53.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:04.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_12/index_677_seed_16516_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:47:23.247 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 22:47:24.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 22:59:16.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_12/index_1914_seed_31883_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 23:00:36.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:36.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:01:48.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_12/index_1914_seed_31884_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:02:08.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 23:02:09.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 23:13:02.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_12/index_707_seed_506_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 23:14:22.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:22.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:32.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_12/index_707_seed_507_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:15:52.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 23:15:53.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 23:26:44.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:50.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_12/index_2627_seed_16082_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 23:28:08.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:28:08.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:15.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_12/index_2627_seed_16083_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:29:38.817 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 23:29:39.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 23:40:27.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:33.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_12/index_292_seed_32693_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 23:41:47.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:47.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:01.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_12/index_292_seed_32694_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:43:25.458 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 23:43:26.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-26 23:54:16.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_12/index_5639_seed_4505_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-26 23:55:35.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:35.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:56:46.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_12/index_5639_seed_4506_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:57:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-26 23:57:09.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.457 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 00:08:40.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:48.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_12/index_3641_seed_32925_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 00:10:06.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:10:06.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:11.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_12/index_3641_seed_32926_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:11:31.975 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 00:11:32.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 00:22:20.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:26.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_12/index_6526_seed_17396_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 00:23:41.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:41.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:24:53.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_12/index_6526_seed_17397_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 00:25:14.490 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:25:14.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 00:25:15.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 00:36:06.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:12.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_12/index_1845_seed_12327_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 00:37:29.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:29.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:38.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_12/index_1845_seed_12328_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:38:58.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 00:38:59.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 00:49:49.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:55.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_12/index_7487_seed_3197_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 00:51:10.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:51:10.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:19.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_12/index_7487_seed_3198_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 00:52:43.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:52:44.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 00:52:44.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 01:03:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_12/index_6071_seed_8811_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 01:04:58.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:58.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:05.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_12/index_6071_seed_8812_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:06:27.724 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 01:06:28.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 01:17:19.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_12/index_7984_seed_21252_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 01:18:41.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:41.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:19:52.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_12/index_7984_seed_21253_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:20:14.640 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 01:20:15.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 01:31:07.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_12/index_1104_seed_10401_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 01:32:28.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:28.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:38.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_12/index_1104_seed_10402_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:33:58.141 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 01:33:59.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 01:44:47.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_12/index_8052_seed_28646_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 01:46:07.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:46:07.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:19.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_12/index_8052_seed_28647_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:47:39.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 01:47:40.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 01:58:27.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:33.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_12/index_6867_seed_7131_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 01:59:46.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:46.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:01.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_12/index_6867_seed_7132_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:01:22.594 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 02:01:23.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 02:12:14.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:15.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_12/index_4498_seed_11023_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 02:13:38.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:38.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:14:47.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_12/index_4498_seed_11024_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:15:08.587 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 02:15:09.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.021 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 02:26:39.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:47.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_12/index_6746_seed_5315_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 02:28:00.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:28:00.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:12.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_12/index_6746_seed_5316_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:29:33.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 02:29:34.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 02:40:23.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_12/index_3944_seed_30076_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 02:41:55.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:55.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:00.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_12/index_3944_seed_30077_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:43:24.598 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 02:43:25.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 02:54:14.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:20.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_12/index_2489_seed_5418_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 02:55:36.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:36.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:56:47.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_12/index_2489_seed_5419_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:57:08.279 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 02:57:09.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 03:08:01.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_12/index_1887_seed_9311_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 03:09:23.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:23.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:31.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_12/index_1887_seed_9312_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:10:49.381 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 03:10:50.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 03:21:41.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:48.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_12/index_1720_seed_32324_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 03:23:02.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:23:02.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:15.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_12/index_1720_seed_32325_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:24:37.786 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 03:24:38.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 03:35:26.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:32.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_12/index_5613_seed_15016_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 03:36:48.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:48.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:02.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_12/index_5613_seed_15017_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:38:24.668 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 03:38:25.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 03:49:14.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:20.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_12/index_3067_seed_23562_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 03:50:34.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:34.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:51:45.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_12/index_3067_seed_23563_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:52:07.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 03:52:08.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 04:03:03.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:09.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_12/index_5391_seed_25674_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 04:04:28.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:28.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:33.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_12/index_5391_seed_25675_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:05:54.105 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 04:05:55.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 04:16:48.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_12/index_8189_seed_24534_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 04:18:08.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:18:08.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:20.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_12/index_8189_seed_24535_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:19:41.823 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 04:19:42.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 04:30:31.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_12/index_8779_seed_13711_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 04:31:49.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:49.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:01.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_12/index_8779_seed_13712_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:33:21.525 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 04:33:22.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.091 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 04:45:58.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_12/index_6441_seed_1666_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 04:47:20.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:20.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:31.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_12/index_6441_seed_1667_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:48:56.024 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 04:48:56.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 04:59:46.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_12/index_5224_seed_7989_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 05:01:05.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:01:05.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:17.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_12/index_5224_seed_7990_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:02:39.195 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 05:02:40.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 05:13:27.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:27.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:33.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_12/index_1605_seed_9323_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 05:14:49.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:49.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:15:56.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_12/index_1605_seed_9324_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:16:16.809 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 05:16:17.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 05:27:07.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_12/index_7303_seed_6555_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 05:28:27.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:27.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:29:40.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_12/index_7303_seed_6556_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:30:00.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 05:30:01.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 05:40:50.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:56.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_12/index_4791_seed_22747_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 05:42:12.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:42:12.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:19.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_12/index_4791_seed_22748_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:43:43.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 05:43:44.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 05:54:35.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_12/index_999_seed_32063_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 05:55:55.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:55.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:07.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_12/index_999_seed_32064_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:57:29.580 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 05:57:30.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 06:08:19.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:25.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_12/index_54_seed_17046_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 06:09:40.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:40.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:10:50.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_12/index_54_seed_17047_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:11:13.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 06:11:14.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 06:22:07.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:07.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_12/index_3738_seed_8865_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 06:23:27.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:27.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:38.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_12/index_3738_seed_8866_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:24:58.868 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 06:24:59.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 06:35:50.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_12/index_9420_seed_19561_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 06:37:13.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:37:13.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:20.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_12/index_9420_seed_19562_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:38:41.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 06:38:42.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 06:49:31.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:36.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_12/index_1078_seed_21830_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 06:50:57.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:57.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:02.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_12/index_1078_seed_21831_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:52:21.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 06:52:21.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.889 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 07:03:49.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_12/index_6744_seed_30154_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 07:05:12.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:05:12.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:20.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_12/index_6744_seed_30155_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:06:39.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 07:06:40.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 07:17:29.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_12/index_2138_seed_6826_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 07:18:53.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:53.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:00.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_12/index_2138_seed_6827_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:20:21.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 07:20:22.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 07:31:16.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_12/index_6274_seed_20847_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 07:32:35.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:35.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:33:49.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_12/index_6274_seed_20848_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:34:11.335 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 07:34:12.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 07:45:07.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:13.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_12/index_8130_seed_16721_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 07:46:28.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:28.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:47:38.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_12/index_8130_seed_16722_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 07:48:01.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:48:01.431 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 07:48:02.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 07:58:52.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_12/index_8050_seed_1419_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 08:00:12.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:00:12.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:23.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_12/index_8050_seed_1420_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:01:45.358 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 08:01:46.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 08:12:38.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_12/index_5871_seed_5581_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 08:13:58.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:58.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:10.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_12/index_5871_seed_5582_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:15:30.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 08:15:31.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 08:26:20.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:26.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_12/index_541_seed_6565_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 08:27:44.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:44.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:28:51.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_12/index_541_seed_6566_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:29:10.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 08:29:11.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 08:40:02.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_12/index_1119_seed_30140_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 08:41:22.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:22.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:35.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_12/index_1119_seed_30141_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:42:57.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 08:42:57.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 08:53:47.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:47.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_12/index_5920_seed_26292_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 08:55:09.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:55:09.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:23.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_12/index_5920_seed_26293_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:56:49.544 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 08:56:50.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 09:07:41.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:48.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_12/index_8993_seed_26285_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 09:09:05.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:09:05.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:13.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_12/index_8993_seed_26286_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:10:36.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 09:10:37.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:00.872 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 09:22:02.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:02.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_12/index_3591_seed_18990_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 09:23:24.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:24.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:34.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_12/index_3591_seed_18991_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:24:54.782 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 09:24:55.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 09:35:43.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:49.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_12/index_7061_seed_29945_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 09:37:05.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:37:05.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:13.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_12/index_7061_seed_29946_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:38:34.451 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 09:38:35.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 09:49:26.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_12/index_9832_seed_30160_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 09:50:49.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:49.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:51:57.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_12/index_9832_seed_30161_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:52:16.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 09:52:18.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 10:03:08.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_12/index_6602_seed_21104_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 10:04:28.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:28.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:39.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_12/index_6602_seed_21105_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:05:59.996 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 10:06:00.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 10:16:53.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_12/index_9366_seed_1670_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 10:18:13.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:18:13.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:24.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_12/index_9366_seed_1671_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:19:45.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 10:19:46.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 10:31:49.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_12/index_1379_seed_7595_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 10:33:15.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:33:15.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:21.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_12/index_1379_seed_7596_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:34:42.919 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 10:34:43.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 10:45:31.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:32.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_12/index_9950_seed_4813_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 10:46:54.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:54.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:03.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_12/index_9950_seed_4814_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:48:26.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 10:48:27.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 10:59:16.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_12/index_7549_seed_11509_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 11:00:36.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:36.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:01:49.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_12/index_7549_seed_11510_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:02:12.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 11:02:13.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.909 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.245 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 11:13:07.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_12/index_5810_seed_20410_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 11:14:30.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:30.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:35.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_12/index_5810_seed_20411_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:15:55.181 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 11:15:56.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 11:26:47.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:53.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_12/index_8781_seed_30533_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 11:28:09.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:28:09.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:21.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_12/index_8781_seed_30534_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:29:41.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 11:29:42.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.094 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 11:41:06.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_12/index_4065_seed_25994_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 11:42:27.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:27.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:37.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_12/index_4065_seed_25995_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:43:58.268 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 11:43:59.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 11:54:48.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_12/index_7359_seed_26215_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 11:56:08.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:56:08.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:20.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_12/index_7359_seed_26216_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:57:44.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 11:57:45.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 12:08:34.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_12/index_6393_seed_10482_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 12:09:59.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:59.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:05.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_12/index_6393_seed_10483_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:11:26.794 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 12:11:27.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 12:22:19.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_12/index_7540_seed_28063_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 12:23:45.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:45.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:24:50.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_12/index_7540_seed_28064_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:25:09.530 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 12:25:10.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 12:36:01.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_12/index_4508_seed_19223_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 12:37:23.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:23.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:33.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_12/index_4508_seed_19224_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:38:52.606 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 12:38:53.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 12:49:44.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:50.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_12/index_9662_seed_26237_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 12:51:04.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:51:04.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:14.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_12/index_9662_seed_26238_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:52:34.135 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 12:52:35.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 13:03:21.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_12/index_5494_seed_7710_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 13:04:42.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:42.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:05:51.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_12/index_5494_seed_7711_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:06:12.755 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 13:06:13.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 13:17:08.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_12/index_3800_seed_25178_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 13:18:28.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:28.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:37.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_12/index_3800_seed_25179_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:19:57.749 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 13:19:58.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 13:30:47.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_12/index_5161_seed_22805_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 13:32:08.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:32:08.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:21.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_12/index_5161_seed_22806_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:33:46.223 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 13:33:47.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 13:44:38.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_12/index_9444_seed_8912_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 13:45:59.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:59.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:09.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_12/index_9444_seed_8913_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:47:30.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 13:47:31.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.110 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 13:58:57.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_12/index_7839_seed_29015_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 14:00:17.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:17.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:30.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_12/index_7839_seed_29016_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:01:50.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 14:01:51.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 14:12:43.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:49.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_12/index_1589_seed_28709_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 14:14:02.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:14:02.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:12.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_12/index_1589_seed_28710_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:15:31.520 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 14:15:32.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 14:26:20.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_12/index_7641_seed_23054_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 14:27:41.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:41.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:28:50.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_12/index_7641_seed_23055_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:29:11.439 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 14:29:12.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 14:40:05.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_12/index_253_seed_10525_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 14:41:32.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:32.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:36.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_12/index_253_seed_10526_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:42:57.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 14:42:58.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 14:53:47.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_12/index_5248_seed_16273_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 14:55:10.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:55:10.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:18.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_12/index_5248_seed_16274_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:56:38.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 14:56:39.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 15:07:26.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:26.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_12/index_2659_seed_23575_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 15:08:49.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:49.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:09:56.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_12/index_2659_seed_23576_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:10:17.193 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 15:10:18.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 15:21:06.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:12.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_12/index_1796_seed_4887_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 15:22:25.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:25.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:37.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_12/index_1796_seed_4888_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:23:57.329 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 15:23:58.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 15:34:47.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_12/index_6772_seed_30405_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 15:36:08.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:36:08.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:20.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_12/index_6772_seed_30406_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:37:43.455 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 15:37:44.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 15:48:32.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_12/index_3240_seed_22226_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 15:49:50.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:50.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:03.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_12/index_3240_seed_22227_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:51:25.017 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 15:51:25.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 16:02:14.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:20.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_12/index_3479_seed_1528_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 16:03:40.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:40.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:04:44.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_12/index_3479_seed_1529_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:05:04.295 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 16:05:05.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.684 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 16:17:43.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_12/index_6301_seed_19601_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 16:19:07.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:19:07.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:17.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_12/index_6301_seed_19602_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:20:37.067 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 16:20:38.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 16:31:26.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_12/index_9466_seed_8661_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 16:32:51.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:51.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:33:58.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_12/index_9466_seed_8662_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:34:17.546 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 16:34:18.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 16:45:07.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_12/index_9391_seed_9236_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 16:46:26.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:26.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:38.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_12/index_9391_seed_9237_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:47:58.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 16:47:59.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 16:58:47.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_12/index_92_seed_28630_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 17:00:08.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:00:08.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:20.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_12/index_92_seed_28631_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:01:41.232 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 17:01:42.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 17:12:31.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_12/index_1835_seed_27714_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 17:13:53.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:53.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:02.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_12/index_1835_seed_27715_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:15:21.535 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 17:15:22.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 17:26:15.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_12/index_8640_seed_33519_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 17:27:34.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:34.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:28:46.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_12/index_8640_seed_33520_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:29:08.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 17:29:09.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 17:40:00.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:06.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_12/index_7740_seed_4042_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 17:41:21.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:21.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:34.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_12/index_7740_seed_4043_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:42:54.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 17:42:55.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 17:53:45.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:51.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_12/index_2878_seed_11191_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 17:55:10.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:55:10.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:15.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_12/index_2878_seed_11192_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:56:37.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 17:56:38.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 18:07:25.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:31.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_12/index_4318_seed_29599_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 18:08:50.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:50.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:01.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_12/index_4318_seed_29600_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:10:22.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 18:10:24.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 18:21:14.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_12/index_2788_seed_4402_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 18:22:41.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:41.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:23:45.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_12/index_2788_seed_4403_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:24:06.984 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 18:24:07.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.229 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 18:35:38.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_12/index_3280_seed_6311_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 18:36:59.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:59.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:09.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_12/index_3280_seed_6312_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:38:31.791 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 18:38:32.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 18:49:21.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_12/index_9460_seed_30763_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 18:50:42.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:42.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:51:55.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_12/index_9460_seed_30764_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:52:15.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 18:52:16.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 19:03:06.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:12.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_12/index_4832_seed_22485_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 19:04:28.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:28.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:36.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_12/index_4832_seed_22486_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:05:55.005 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 19:05:55.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 19:16:45.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_12/index_2067_seed_29199_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 19:18:08.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:18:08.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:21.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_12/index_2067_seed_29200_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:19:42.104 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 19:19:42.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 19:30:30.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_12/index_6156_seed_30084_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 19:31:53.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:53.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:04.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_12/index_6156_seed_30085_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:33:24.294 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 19:33:25.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 19:44:14.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_12/index_997_seed_21136_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 19:45:33.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:33.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:46:44.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_12/index_997_seed_21137_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:47:09.151 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 19:47:10.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 19:58:02.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_12/index_7526_seed_29815_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 19:59:23.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:23.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:33.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_12/index_7526_seed_29816_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:00:53.200 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 20:00:53.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 20:11:43.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_12/index_9553_seed_18416_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 20:13:03.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:13:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:15.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_12/index_9553_seed_18417_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:14:38.280 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 20:14:39.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 20:25:26.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_12/index_1783_seed_19068_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 20:26:51.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:51.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:27:58.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_12/index_1783_seed_19069_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:28:18.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 20:28:19.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 20:39:06.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_12/index_7902_seed_22617_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 20:40:29.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:29.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:41:38.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_12/index_7902_seed_22618_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:42:03.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 20:42:04.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.763 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 20:53:29.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:35.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_12/index_7699_seed_18108_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 20:54:53.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:53.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:55:59.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_12/index_7699_seed_18109_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:56:20.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 20:56:21.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 21:07:10.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:16.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_12/index_2123_seed_15493_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 21:08:33.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:33.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:09:41.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_12/index_2123_seed_15494_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:10:06.188 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 21:10:07.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 21:21:01.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_12/index_3616_seed_23083_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 21:22:23.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:23.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:31.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_12/index_3616_seed_23084_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:23:51.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 21:23:52.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 21:34:42.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:48.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_12/index_8502_seed_15656_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 21:36:04.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:36:04.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:12.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_12/index_8502_seed_15657_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:37:33.776 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 21:37:34.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 21:48:23.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_12/index_8175_seed_5844_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 21:49:44.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:44.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:50:54.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_12/index_8175_seed_5845_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 21:51:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:51:15.143 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 21:51:15.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 22:03:21.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_12/index_423_seed_26955_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 22:04:40.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:40.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:05:54.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_12/index_423_seed_26956_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:06:16.161 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 22:06:16.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 22:17:06.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_12/index_8675_seed_11560_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 22:18:27.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:27.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:37.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_12/index_8675_seed_11561_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 22:19:56.152 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:19:56.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 22:19:57.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 22:30:47.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_12/index_5989_seed_22419_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 22:32:11.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:32:11.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:21.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_12/index_5989_seed_22420_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:33:40.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 22:33:41.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 22:44:30.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:36.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_12/index_8379_seed_3873_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 22:45:53.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:53.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:00.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_12/index_8379_seed_3874_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:47:17.416 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 22:47:18.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 22:58:06.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_12/index_5783_seed_24219_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 22:59:28.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:28.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:00:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_12/index_5783_seed_24220_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:01:00.699 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 23:01:01.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.551 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 23:12:29.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_12/index_6212_seed_16383_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 23:13:52.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:52.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:01.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_12/index_6212_seed_16384_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:15:22.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 23:15:23.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 23:26:15.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_12/index_5436_seed_13660_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 23:27:36.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:36.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:28:45.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_12/index_5436_seed_13661_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:29:06.049 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 23:29:06.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 23:40:02.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_12/index_1085_seed_2393_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 23:41:22.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:22.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:42:36.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_12/index_1085_seed_2394_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:43:00.919 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 23:43:01.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-27 23:53:51.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_12/index_4182_seed_25125_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-27 23:55:14.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:55:14.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:22.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_12/index_4182_seed_25126_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:56:43.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-27 23:56:44.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 00:07:32.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:33.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:38.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_12/index_5170_seed_8513_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 00:08:53.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:53.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:08.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_12/index_5170_seed_8514_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:10:30.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 00:10:31.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 00:21:21.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_12/index_3215_seed_12328_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 00:22:42.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:42.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:23:51.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_12/index_3215_seed_12329_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:24:11.487 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 00:24:12.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 00:35:07.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:13.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_12/index_9269_seed_28388_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 00:36:27.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:27.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:37:45.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_12/index_9269_seed_28389_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:38:09.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 00:38:10.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 00:49:02.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:08.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_12/index_7053_seed_7160_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 00:50:24.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:24.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:36.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_12/index_7053_seed_7161_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:51:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 00:51:57.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 01:02:48.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:54.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_12/index_6098_seed_10673_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 01:04:10.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:04:10.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:21.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_12/index_6098_seed_10674_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:05:43.950 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 01:05:44.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 01:16:33.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_12/index_3813_seed_25322_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 01:17:55.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:55.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:11.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_12/index_3813_seed_25323_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:19:30.583 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 01:19:31.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.670 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 01:30:58.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_12/index_1011_seed_12178_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 01:32:24.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:24.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:29.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_12/index_1011_seed_12179_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:33:51.857 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 01:33:52.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 01:44:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:50.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_12/index_8597_seed_6386_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 01:46:06.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:46:06.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:16.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_12/index_8597_seed_6387_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:47:36.883 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 01:47:37.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 01:58:27.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:33.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_12/index_8225_seed_20148_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 01:59:46.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:46.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:00:59.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_12/index_8225_seed_20149_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:01:17.853 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 02:01:18.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 02:12:08.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:14.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_12/index_9653_seed_1589_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 02:13:33.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:33.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:14:41.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_12/index_9653_seed_1590_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:15:02.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 02:15:04.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 02:25:53.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:59.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_12/index_2327_seed_9317_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 02:27:19.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:27:19.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:25.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_12/index_2327_seed_9318_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:28:44.786 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 02:28:45.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 02:39:38.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:44.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_12/index_1079_seed_23176_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 02:41:02.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:41:02.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:08.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_12/index_1079_seed_23177_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:42:30.273 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 02:42:31.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 02:53:20.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_12/index_180_seed_15561_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 02:54:41.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:41.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:55:52.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_12/index_180_seed_15562_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:56:12.573 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 02:56:13.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-28 02:58:07.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 03:07:08.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:14.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_12/index_1094_seed_17884_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 03:08:29.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:29.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:09:38.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_12/index_1094_seed_17885_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:10:01.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 03:10:03.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 03:20:53.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_12/index_905_seed_14225_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 03:22:15.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:22:15.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:26.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_12/index_905_seed_14226_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:23:47.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 03:23:49.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 03:34:40.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_12/index_1122_seed_8559_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 03:36:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:36:03.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:12.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_12/index_1122_seed_8560_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:37:34.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 03:37:35.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.327 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 03:50:13.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_12/index_132_seed_6730_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 03:51:35.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:35.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:52:46.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_12/index_132_seed_6731_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:53:05.855 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 03:53:06.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.122 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 04:04:03.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_12/index_1207_seed_21579_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 04:05:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:23.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:33.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_12/index_1207_seed_21580_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:06:55.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 04:06:56.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 04:17:48.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_12/index_8382_seed_32193_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 04:19:12.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:19:12.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:20.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_12/index_8382_seed_32194_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:20:45.839 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 04:20:46.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 04:31:40.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_12/index_3423_seed_5095_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 04:32:58.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:58.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:09.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_12/index_3423_seed_5096_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:34:28.738 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 04:34:29.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 04:45:20.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:26.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_12/index_6043_seed_3445_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 04:46:44.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:44.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:47:50.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_12/index_6043_seed_3446_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:48:09.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 04:48:11.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 04:59:03.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_12/index_2589_seed_30833_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 05:00:25.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:25.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:32.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_12/index_2589_seed_30834_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:01:51.670 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 05:01:52.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 05:12:44.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:45.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_12/index_8043_seed_23226_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 05:14:05.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:14:05.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:15.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_12/index_8043_seed_23227_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:15:34.178 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 05:15:35.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 05:26:25.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_12/index_539_seed_22503_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 05:27:47.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:47.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:28:56.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_12/index_539_seed_22504_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:29:21.195 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 05:29:22.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 05:40:15.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_12/index_1823_seed_5346_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 05:41:35.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:35.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:42:47.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_12/index_1823_seed_5347_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:43:09.995 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 05:43:10.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 05:54:06.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:07.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_12/index_7541_seed_25257_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 05:55:33.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:33.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:38.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_12/index_7541_seed_25258_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:56:59.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 05:57:00.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.460 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 06:08:27.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_12/index_5645_seed_29952_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 06:09:48.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:48.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:10:58.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_12/index_5645_seed_29953_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:11:20.375 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 06:11:21.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 06:22:11.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_12/index_915_seed_26645_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 06:23:31.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:31.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:24:42.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_12/index_915_seed_26646_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:25:06.233 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 06:25:07.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 06:36:02.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_12/index_2774_seed_1805_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 06:37:23.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:23.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:32.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_12/index_2774_seed_1806_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:38:52.704 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 06:38:53.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 06:49:45.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:51.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_12/index_627_seed_26223_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 06:51:06.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:51:06.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:16.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_12/index_627_seed_26224_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:52:36.969 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 06:52:37.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 07:03:27.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:33.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_12/index_2470_seed_23679_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 07:04:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:49.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:05:59.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_12/index_2470_seed_23680_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:06:20.768 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 07:06:21.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 07:17:12.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_12/index_5332_seed_27775_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 07:18:33.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:33.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:19:42.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_12/index_5332_seed_27776_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:20:03.223 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 07:20:04.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 07:30:53.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:59.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_12/index_2581_seed_11487_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 07:32:17.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:32:17.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:26.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_12/index_2581_seed_11488_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:33:47.191 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 07:33:48.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 07:44:40.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:46.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_12/index_4185_seed_3329_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 07:46:00.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:46:00.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:12.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_12/index_4185_seed_3330_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:47:32.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 07:47:33.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 07:58:21.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_12/index_5911_seed_15191_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 07:59:46.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:46.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:00:52.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_12/index_5911_seed_15192_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:01:15.774 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 08:01:16.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 08:12:08.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_12/index_4586_seed_11075_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 08:13:32.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:32.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:14:39.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_12/index_4586_seed_11076_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:15:02.413 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 08:15:03.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.415 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 08:26:31.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_12/index_3785_seed_17031_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 08:27:52.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:52.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:02.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_12/index_3785_seed_17032_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:29:27.132 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 08:29:27.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 08:40:22.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_12/index_2790_seed_19715_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 08:41:48.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:48.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:42:53.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_12/index_2790_seed_19716_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:43:13.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 08:43:14.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 08:54:07.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_12/index_833_seed_29172_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 08:55:28.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:28.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_12/index_833_seed_29173_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:56:59.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 08:57:00.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 09:07:50.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:51.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_12/index_3968_seed_27_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 09:09:10.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:09:10.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:21.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_12/index_3968_seed_28_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:10:43.118 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 09:10:43.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 09:21:32.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_12/index_5599_seed_24572_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 09:22:53.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:53.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:04.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_12/index_5599_seed_24573_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:24:27.683 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 09:24:28.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 09:36:35.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_12/index_669_seed_17781_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 09:37:54.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:54.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:05.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_12/index_669_seed_17782_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 09:39:26.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:39:26.869 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 09:39:27.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 09:50:22.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:28.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_12/index_4942_seed_11388_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 09:51:47.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:47.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:52:55.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_12/index_4942_seed_11389_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:53:24.108 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 09:53:25.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 10:04:16.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_12/index_3011_seed_1199_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 10:05:38.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:38.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:06:50.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_12/index_3011_seed_1200_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:07:11.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 10:07:11.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 10:18:09.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_12/index_7613_seed_1167_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 10:19:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:20:40.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_12/index_7613_seed_1168_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:21:00.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 10:21:02.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 10:31:54.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_12/index_6374_seed_16295_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 10:33:14.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:33:14.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:26.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_12/index_6374_seed_16296_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:34:53.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 10:34:54.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.709 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 10:46:20.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_12/index_137_seed_11715_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 10:47:45.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:45.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:48:51.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_12/index_137_seed_11716_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:49:12.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 10:49:13.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.442 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 11:00:08.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_12/index_424_seed_26192_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 11:01:27.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:27.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:39.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_12/index_424_seed_26193_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:02:59.634 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 11:03:00.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 11:13:51.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:57.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_12/index_2754_seed_3284_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 11:15:11.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:15:11.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:21.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_12/index_2754_seed_3285_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:16:42.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 11:16:43.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 11:27:32.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:38.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_12/index_1318_seed_20778_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 11:28:53.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:53.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:02.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_12/index_1318_seed_20779_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:30:22.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 11:30:23.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 11:41:15.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:21.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_12/index_7094_seed_16974_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 11:42:37.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:37.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:43:47.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_12/index_7094_seed_16975_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:44:07.739 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 11:44:08.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 11:55:03.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:09.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_12/index_3666_seed_23407_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 11:56:29.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:29.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:34.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_12/index_3666_seed_23408_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:57:56.497 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 11:57:57.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 12:08:48.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_12/index_6765_seed_30384_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 12:10:12.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:10:12.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:25.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_12/index_6765_seed_30385_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:11:47.114 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 12:11:48.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 12:22:40.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:46.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_12/index_1258_seed_18357_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 12:24:01.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:24:01.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:13.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_12/index_1258_seed_18358_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:25:34.015 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 12:25:34.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 12:36:24.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_12/index_8030_seed_28554_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 12:37:42.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:42.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:38:55.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_12/index_8030_seed_28555_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:39:16.520 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 12:39:17.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 12:50:08.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_12/index_8943_seed_15114_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 12:51:32.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:32.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:52:41.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_12/index_8943_seed_15115_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:53:04.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 12:53:04.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.899 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 13:04:33.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_12/index_7725_seed_25572_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 13:05:55.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:55.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:07.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_12/index_7725_seed_25573_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 13:07:33.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:07:33.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 13:07:34.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 13:18:22.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_12/index_2036_seed_8000_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 13:19:48.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:48.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:20:53.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_12/index_2036_seed_8001_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:21:13.437 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 13:21:14.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 13:32:07.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_12/index_4757_seed_14435_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 13:33:28.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:28.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:38.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_12/index_4757_seed_14436_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:34:59.762 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 13:35:00.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 13:45:49.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:49.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_12/index_4306_seed_5848_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 13:47:09.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:47:09.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:22.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_12/index_4306_seed_5849_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:48:41.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 13:48:42.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 13:59:30.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:30.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_12/index_4211_seed_30515_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 14:00:51.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:51.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:00.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_12/index_4211_seed_30516_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:02:20.382 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 14:02:21.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 14:13:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:14.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_12/index_9012_seed_9743_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 14:14:30.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:30.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:15:42.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_12/index_9012_seed_9744_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:16:08.634 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 14:16:09.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 14:27:02.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_12/index_4672_seed_7616_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 14:28:21.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:21.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:31.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_12/index_4672_seed_7617_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:29:50.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 14:29:51.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 14:40:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_12/index_6130_seed_4671_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 14:42:07.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:42:07.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:15.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_12/index_6130_seed_4672_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:43:35.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 14:43:35.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 14:54:25.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_12/index_8010_seed_29106_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 14:55:47.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:47.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:56:55.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_12/index_8010_seed_29107_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:57:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 14:57:15.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 15:08:07.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_12/index_6799_seed_26907_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 15:09:29.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:29.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:10:39.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_12/index_6799_seed_26908_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:11:01.171 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 15:11:02.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.091 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 15:23:36.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_12/index_8133_seed_24970_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 15:25:00.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:25:00.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:08.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_12/index_8133_seed_24971_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:26:31.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 15:26:31.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 15:37:21.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_12/index_7242_seed_28361_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 15:38:46.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:46.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:39:52.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_12/index_7242_seed_28362_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:40:14.111 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 15:40:15.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 15:51:08.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:14.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_12/index_1479_seed_32931_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 15:52:26.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:26.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:53:40.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_12/index_1479_seed_32932_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:54:03.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 15:54:05.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 16:04:54.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:06:00.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_12/index_2707_seed_12808_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 16:06:14.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:06:14.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:24.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_12/index_2707_seed_12809_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:07:49.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 16:07:50.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 16:18:41.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:41.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_12/index_5251_seed_31157_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 16:19:59.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:59.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:12.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_12/index_5251_seed_31158_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:21:33.498 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 16:21:34.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 16:32:23.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_12/index_9755_seed_4683_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 16:33:43.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:34:53.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_12/index_9755_seed_4684_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:35:13.567 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 16:35:14.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.524 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 16:46:07.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_12/index_4876_seed_14299_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 16:47:26.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:26.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:39.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_12/index_4876_seed_14300_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:48:59.496 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 16:49:00.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 16:59:50.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_12/index_790_seed_19721_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 17:01:12.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:01:12.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:21.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_12/index_790_seed_19722_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 17:02:43.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:02:43.833 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 17:02:44.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 17:13:34.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_12/index_2092_seed_27848_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 17:14:54.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:54.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:08.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_12/index_2092_seed_27849_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:16:31.131 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 17:16:32.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 17:27:21.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_12/index_9799_seed_6855_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 17:28:43.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:43.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:29:53.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_12/index_9799_seed_6856_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:30:17.796 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 17:30:18.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.945 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 17:41:44.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_12/index_3506_seed_27580_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 17:43:05.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:43:05.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:21.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_12/index_3506_seed_27581_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:44:42.558 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 17:44:43.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 17:55:31.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:32.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:37.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_12/index_8798_seed_30334_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 17:56:51.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:51.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:02.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_12/index_8798_seed_30335_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:58:24.755 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 17:58:25.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 18:09:15.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_12/index_3000_seed_18303_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 18:10:36.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:37.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:11:49.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_12/index_3000_seed_18304_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:12:09.313 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 18:12:10.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 18:23:03.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_12/index_2640_seed_1134_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 18:24:20.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:20.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:33.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_12/index_2640_seed_1135_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:25:53.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 18:25:54.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 18:36:44.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_12/index_646_seed_28682_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 18:38:06.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:38:06.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:16.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_12/index_646_seed_28683_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:39:38.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 18:39:39.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 18:50:26.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:32.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_12/index_4814_seed_6748_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 18:51:46.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:46.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:52:59.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_12/index_4814_seed_6749_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:53:21.465 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 18:53:22.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 19:04:13.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_12/index_9792_seed_33191_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 19:05:37.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:37.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:06:45.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_12/index_9792_seed_33192_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:07:07.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 19:07:08.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.142 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 19:18:01.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_12/index_5771_seed_18633_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 19:19:19.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:19.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:30.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_12/index_5771_seed_18634_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:20:52.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 19:20:53.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 19:31:44.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_12/index_5024_seed_14613_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 19:33:05.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:33:05.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:16.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_12/index_5024_seed_14614_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 19:34:36.662 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:34:36.818 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 19:34:37.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 19:45:24.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_12/index_4444_seed_32509_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 19:46:46.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:46.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:47:56.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_12/index_4444_seed_32510_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:48:17.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 19:48:18.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.888 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 19:59:44.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:50.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_12/index_44_seed_19111_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 20:01:04.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:01:04.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:15.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_12/index_44_seed_19112_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:02:34.520 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 20:02:35.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 20:13:24.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_12/index_4387_seed_23845_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 20:14:50.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:50.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:15:55.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_12/index_4387_seed_23846_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:16:15.201 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 20:16:16.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 20:27:07.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_12/index_9474_seed_21671_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 20:28:31.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:31.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:38.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_12/index_9474_seed_21672_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:29:58.628 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 20:29:59.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 20:40:49.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_12/index_8891_seed_4675_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 20:42:10.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:42:10.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:22.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_12/index_8891_seed_4676_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:43:45.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 20:43:46.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 20:54:36.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:42.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_12/index_6905_seed_21521_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 20:55:56.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:56.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:09.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_12/index_6905_seed_21522_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:57:33.593 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 20:57:34.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 21:09:41.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:48.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_12/index_7353_seed_12553_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 21:11:01.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:11:01.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:14.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_12/index_7353_seed_12554_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:12:34.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 21:12:35.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 21:23:23.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_12/index_4650_seed_22605_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 21:24:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:43.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:25:53.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_12/index_4650_seed_22606_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:26:15.011 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 21:26:15.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 21:37:07.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_12/index_4927_seed_6091_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 21:38:31.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:31.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:36.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_12/index_4927_seed_6092_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:39:57.784 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 21:39:58.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 21:50:48.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_12/index_3751_seed_9066_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 21:52:09.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:52:09.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:21.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_12/index_3751_seed_9067_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:53:42.487 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 21:53:43.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 22:04:31.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:31.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_12/index_4886_seed_8166_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 22:05:53.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:53.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:04.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_12/index_4886_seed_8167_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:07:25.908 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 22:07:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.393 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 22:18:53.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_12/index_808_seed_12288_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 22:20:14.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:20:14.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:24.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_12/index_808_seed_12289_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:21:46.281 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 22:21:47.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 22:32:40.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_12/index_4622_seed_30851_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 22:34:02.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:34:02.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:11.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_12/index_4622_seed_30852_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:35:34.441 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 22:35:35.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 22:46:24.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_12/index_4523_seed_5666_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 22:47:43.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:43.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:48:56.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_12/index_4523_seed_5667_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:49:17.645 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 22:49:18.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:06.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 23:00:07.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_12/index_418_seed_17388_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 23:01:28.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:28.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:38.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_12/index_418_seed_17389_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:02:58.421 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 23:02:59.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 23:13:49.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:50.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:55.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_12/index_4234_seed_29703_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 23:15:11.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:15:11.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:20.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_12/index_4234_seed_29704_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:16:43.582 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 23:16:44.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 23:27:32.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:38.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_12/index_463_seed_32735_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 23:28:56.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:56.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:02.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_12/index_463_seed_32736_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:30:22.236 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 23:30:23.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 23:41:14.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_12/index_822_seed_7396_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 23:42:32.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:32.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:43:44.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_12/index_822_seed_7397_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 23:44:05.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:44:05.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 23:44:06.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-28 23:55:00.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_12/index_2826_seed_31015_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-28 23:56:20.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:20.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:29.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_12/index_2826_seed_31016_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:57:49.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-28 23:57:49.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-29 00:08:39.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:45.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_12/index_9960_seed_14882_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-29 00:09:59.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:59.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:08.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_12/index_9960_seed_14883_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:11:28.258 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-29 00:11:29.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-29 00:22:19.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:25.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_12/index_7222_seed_2584_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-29 00:23:39.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:39.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:24:52.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_12/index_7222_seed_2585_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:25:12.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-29 00:25:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:40.896 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-29 00:36:43.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:49.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_12/index_9330_seed_31144_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-29 00:38:07.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:38:07.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:15.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_12/index_9330_seed_31145_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:39:38.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-29 00:39:39.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-29 00:50:27.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:33.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_12/index_1412_seed_13467_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-29 00:51:48.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:48.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:52:56.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_12/index_1412_seed_13468_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:53:18.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-29 00:53:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.326 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 48, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 48, dp_rank 24: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=0 +2026-06-29 01:04:09.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:15.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_12/index_3710_seed_29955_dp_24_rank_48_sample_0_slot_0.mp4 +2026-06-29 01:05:29.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 48 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:29.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 48 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:06:41.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 48: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_12/index_3710_seed_29956_dp_24_rank_48_sample_1_slot_1.mp4 +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 48: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 01:07:03.013 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1101 - [GatherRewards] Rank 48 saved 4 entries for group 12 +2026-06-29 01:07:04.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 48: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 48: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 48, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_50.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_50.log new file mode 100644 index 0000000000000000000000000000000000000000..068dc65448f50781332b49a8349faf6addf75031 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_50.log @@ -0,0 +1,6079 @@ +2026-05-29 20:45:42.519 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-29 20:45:42.519 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-29 20:45:43.208 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:45:43.208 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 20:51:46.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26132_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 20:53:31.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:31.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:38.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26133_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 20:55:17.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 50, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 21:06:57.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:04.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26457_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 21:08:19.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:19.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:09:39.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26458_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:10:01.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 21:21:17.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1346_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 21:22:38.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:38.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:23:53.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1347_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:24:15.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 21:35:35.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1784_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 21:36:58.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:58.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:14.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1785_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:38:36.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 21:49:56.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26286_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 21:51:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:21.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:52:35.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26287_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:53:01.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 22:04:18.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28029_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 22:05:41.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:41.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:06:52.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28030_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:07:15.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 22:18:35.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:42.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19488_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 22:19:57.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:57.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:14.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19489_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:21:38.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 22:32:58.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:58.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16831_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 22:34:22.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:22.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:36.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16832_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:36:01.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 22:47:20.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6182_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 22:48:42.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:42.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:49:56.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6183_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:50:20.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 23:01:44.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29913_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 23:03:08.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:03:08.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:23.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29914_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:04:49.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:53.324 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 23:16:57.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27367_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 23:18:22.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:22.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:35.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27368_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:19:57.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 23:31:14.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8791_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 23:32:33.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:33.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:33:52.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8792_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:34:15.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 23:45:29.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:35.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7436_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-29 23:46:48.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:48.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:09.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7437_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:48:31.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-29 23:59:50.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29113_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 00:01:14.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:01:14.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:26.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29114_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:02:55.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 00:14:13.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1987_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 00:15:39.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:39.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:16:51.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1988_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:17:18.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 00:28:36.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:46.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32332_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 00:30:01.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:30:01.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:16.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32333_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:31:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 00:42:58.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20821_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 00:44:20.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:20.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:45:38.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20822_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:46:02.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 00:57:15.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9955_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 00:58:41.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:41.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:59:51.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9956_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:00:16.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 01:11:38.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:45.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6673_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 01:12:59.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:59.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:18.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6674_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:14:44.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 01:26:05.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:14.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7233_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 01:27:30.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:30.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:28:42.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7234_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:29:04.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.568 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 01:41:19.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:27.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18723_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 01:42:42.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:42.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:43:56.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18724_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:44:18.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 01:55:38.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9950_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 01:56:58.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:58.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:14.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9951_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:58:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 02:09:58.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21359_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 02:11:19.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:19.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:12:39.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21360_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:13:02.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 02:24:23.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16560_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 02:25:43.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:43.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:05.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16561_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:27:33.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 02:38:56.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_27_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 02:40:20.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:20.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:35.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_28_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:41:58.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 02:54:31.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15324_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 02:56:02.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:56:02.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:08.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15325_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:57:31.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 03:08:50.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9663_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 03:10:10.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:10:10.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:27.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9664_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:11:53.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 03:23:09.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:17.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20499_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 03:24:37.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:37.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:25:46.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20500_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 03:26:07.442 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:26:08.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 03:37:26.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14887_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 03:38:51.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:51.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:07.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14888_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:40:39.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 03:51:59.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28205_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 03:53:23.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:23.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:54:38.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28206_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:55:04.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.395 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 04:07:19.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:28.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17148_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 04:08:47.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:47.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:01.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17149_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:10:28.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 04:21:48.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15659_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 04:23:10.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:23:10.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:28.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15660_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:24:54.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 04:36:12.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10429_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 04:37:34.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:34.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:38:50.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10430_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:39:16.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.615 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 04:50:41.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:48.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_1999_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 04:52:03.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:52:03.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:19.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_2000_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:53:44.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 05:05:07.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3068_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 05:06:30.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:30.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:07:43.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3069_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:08:12.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 05:19:25.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:32.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12882_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 05:20:47.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:47.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:21:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12883_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:22:20.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 05:33:38.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:46.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12611_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 05:35:04.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:35:04.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:15.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12612_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:36:37.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 05:47:59.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8703_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 05:49:18.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:18.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:35.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8704_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:50:58.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 06:02:21.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:30.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2227_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 06:03:45.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:45.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:04.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2228_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:05:33.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 06:16:55.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6627_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 06:18:20.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:20.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:31.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6628_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 06:19:58.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:19:59.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:12.074 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 06:32:17.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:26.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17612_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 06:33:41.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:41.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:34:55.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17613_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:35:16.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 06:46:36.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21422_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 06:48:00.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:48:00.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:15.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21423_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:49:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 07:00:59.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26057_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 07:02:27.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:27.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:37.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26058_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:04:00.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 07:15:21.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:29.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15760_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 07:16:45.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:45.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:06.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15761_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:18:41.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 07:30:08.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:17.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18245_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 07:31:31.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:32:48.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18246_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:33:11.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 07:44:30.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25139_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 07:45:54.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:54.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:04.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25140_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:47:30.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 07:59:00.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:08.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8944_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 08:00:24.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:24.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:01:41.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8945_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:02:08.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 08:13:34.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21476_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 08:14:59.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:59.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:17.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21477_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:16:44.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 08:28:15.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_97_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 08:29:41.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:41.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:30:55.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_98_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:31:23.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 08:42:53.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7488_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 08:44:17.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:17.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:45:35.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7489_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:46:01.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:24.855 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 08:59:29.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5951_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 09:00:57.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:57.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:11.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5952_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:02:33.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 09:14:05.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:12.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18298_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 09:15:30.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:30.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:16:52.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18299_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:17:17.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 09:28:42.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18668_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 09:30:07.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:30:07.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:27.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18669_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:31:54.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 09:43:17.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:26.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17710_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 09:44:41.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:41.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:02.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17711_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:46:24.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 09:57:46.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25592_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 09:59:09.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:59:09.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:23.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25593_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:00:52.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 10:12:07.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14040_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 10:13:31.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:31.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:14:44.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14041_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:15:06.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 10:26:24.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6820_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 10:27:49.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:49.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:02.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6821_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:29:29.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 10:40:46.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14362_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 10:42:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:42:09.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:28.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14363_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:43:55.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 10:55:08.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11085_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 10:56:31.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:31.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:57:52.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11086_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:58:23.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 11:09:40.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9324_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 11:11:05.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:11:05.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:20.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9325_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:12:43.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:03.434 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 11:25:08.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24003_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 11:26:34.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:34.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:27:53.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24004_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 11:28:20.076 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:28:21.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 11:39:42.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7438_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 11:41:05.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:41:05.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:19.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7439_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:42:42.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 11:54:06.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:14.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_960_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 11:55:29.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:29.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:56:42.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_961_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:57:06.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 12:08:30.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11097_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 12:09:54.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:54.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:07.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11098_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:11:30.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 12:22:53.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13339_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 12:24:23.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:23.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:30.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13340_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:25:55.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 12:37:12.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24634_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 12:38:35.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:35.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:39:47.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24635_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 12:40:10.052 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:40:11.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 12:51:30.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31475_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 12:52:56.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:56.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:11.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31476_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:54:32.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 13:05:59.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17026_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 13:07:19.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:19.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:08:38.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17027_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:09:04.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 13:20:24.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:32.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6381_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 13:21:46.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:46.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:00.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6382_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:23:30.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 13:34:48.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33216_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 13:36:11.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:36:11.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:25.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33217_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:37:48.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.846 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 13:50:02.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16435_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 13:51:26.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:26.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:52:48.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16436_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:53:11.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 14:04:33.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:44.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14142_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 14:05:56.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:56.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:19.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14143_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:07:59.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 14:19:43.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1068_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 14:21:10.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:21:10.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:22:39.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1069_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:23:13.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 14:34:38.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:46.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16677_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 14:36:00.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:36:00.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:16.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16678_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:37:54.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 14:49:20.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3083_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 14:50:43.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:43.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:00.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3084_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:52:26.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 15:05:01.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:10.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14049_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 15:06:25.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:25.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:07:40.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14050_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:08:04.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 15:19:35.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:43.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10071_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 15:21:02.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:21:02.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:14.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10072_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:22:38.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 15:34:11.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21021_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 15:35:36.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:36:53.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21022_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:37:19.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 15:48:48.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8838_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 15:50:12.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:50:12.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:25.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8839_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:51:51.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 16:03:17.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14210_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 16:04:39.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:39.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:05:57.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14211_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:06:20.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.250 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 16:18:48.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32870_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 16:20:20.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:20.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:21:33.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32871_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:22:05.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 16:33:33.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:33.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:43.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7824_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 16:34:56.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:57.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:16.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7825_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:36:40.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 16:48:15.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:24.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5779_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 16:49:39.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:39.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:50:53.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5780_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:51:16.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 17:02:55.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:04.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11198_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 17:04:21.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:21.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:33.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11199_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:05:57.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 17:17:35.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20780_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 17:19:03.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:19:03.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:15.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20781_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:20:38.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 17:32:10.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29859_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 17:33:32.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:32.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:00.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29860_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:35:26.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 17:46:54.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20793_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 17:48:22.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:22.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:49:38.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20794_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:50:02.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 18:01:35.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12960_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 18:02:58.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:58.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:17.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12961_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:04:40.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 18:16:18.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32888_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 18:17:41.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:41.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:00.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32889_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:19:32.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 18:31:18.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:30.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9543_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 18:32:44.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:44.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:14.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9544_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 18:34:45.308 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:34:46.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:18.255 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 18:47:24.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11842_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 18:48:48.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:48.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:12.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11843_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 18:50:38.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:50:39.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 19:02:13.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19513_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 19:03:41.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:41.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:01.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19514_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:05:34.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 19:17:12.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:22.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29115_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 19:18:37.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:37.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:19:59.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29116_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:20:26.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:53.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 19:31:55.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18915_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 19:33:22.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:22.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:34:55.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18916_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:35:24.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 19:47:03.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24558_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 19:48:27.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:27.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:49:47.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24559_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:50:11.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 20:01:46.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:57.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12257_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 20:03:13.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:03:13.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:29.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12258_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:04:56.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 20:16:29.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:41.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18521_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 20:17:54.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:54.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:19.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18522_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:19:47.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 20:31:20.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:29.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15433_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 20:32:44.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:44.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:00.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15434_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:34:24.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 20:46:03.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16701_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 20:47:28.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:28.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:48:49.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16702_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:49:15.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 21:00:48.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25197_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 21:02:13.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:02:13.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:31.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25198_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:03:58.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.216 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 21:17:31.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:41.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10951_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 21:18:56.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:56.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:12.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10952_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:20:37.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 21:32:08.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:49:21.311 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-30 21:49:21.312 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-30 21:49:22.012 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:49:22.012 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 21:55:22.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:49.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26132_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 21:57:05.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:57:05.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:18.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26133_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:59:08.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 50, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 22:10:52.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:12:01.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26457_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 22:12:14.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:14.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:13:35.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26458_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:14:03.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 22:25:33.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:42.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1346_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 22:26:56.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:56.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1347_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:28:37.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 22:40:03.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:04.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1784_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 22:41:26.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:26.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:42:45.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1785_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:43:10.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 22:54:38.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:38.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:47.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26286_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 22:56:01.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:56:01.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:25.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26287_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:57:50.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 23:09:20.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28029_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 23:10:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:44.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:11.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28030_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:12:36.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 23:24:05.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19488_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 23:25:27.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:27.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:26:47.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19489_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:27:14.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 23:38:46.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:55.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16831_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 23:40:11.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:40:11.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:25.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16832_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:41:48.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-30 23:53:10.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6182_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-30 23:54:33.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:33.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:55:52.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6183_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:56:15.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.211 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 00:07:41.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29913_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 00:09:09.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:09:09.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:24.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29914_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:10:46.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:54.307 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 00:22:59.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27367_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 00:24:26.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:26.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:25:40.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27368_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:26:05.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 00:37:30.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8791_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 00:38:55.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:55.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:18.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8792_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:40:46.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 00:52:07.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7436_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 00:53:28.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:28.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:54:49.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7437_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:55:21.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 01:06:44.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29113_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 01:08:11.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:08:11.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:26.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29114_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:09:52.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 01:21:16.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1987_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 01:22:44.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:44.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:23:55.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1988_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:24:20.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 01:35:36.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32332_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 01:36:57.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:57.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:15.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32333_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:38:38.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 01:50:03.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20821_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 01:51:27.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:27.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:52:42.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20822_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 01:53:05.979 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:53:06.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 02:04:34.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9955_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 02:06:00.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:06:00.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:13.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9956_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:07:39.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 50, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-05-31 02:19:01.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6673_dp_25_rank_50_sample_2_slot_0.mp4 +2026-05-31 02:20:25.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:25.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:21:38.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6674_dp_25_rank_50_sample_3_slot_1.mp4 +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:22:02.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:03:35.311 | INFO | __main__:__init__:946 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-06-25 18:03:35.311 | INFO | __main__:__init__:947 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-06-25 18:03:35.984 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:03:35.984 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 18:04:09.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26132_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 18:05:47.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:47.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:06:56.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_12/index_6934_seed_26133_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:07:32.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.293 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 50, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 18:18:41.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:47.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26457_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 18:20:01.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:20:01.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:12.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_12/index_5051_seed_26458_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:21:35.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.711 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 18:32:26.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1346_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 18:33:47.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:47.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:34:54.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_12/index_7027_seed_1347_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:35:15.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 18:46:08.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1784_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 18:47:27.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:27.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:48:38.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_12/index_6106_seed_1785_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:49:03.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 19:00:02.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:02.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26286_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 19:01:21.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:21.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:34.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_12/index_4135_seed_26287_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:02:57.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 19:13:49.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28029_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 19:15:08.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:15:08.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:20.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_12/index_181_seed_28030_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:16:42.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 19:27:32.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19488_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 19:28:56.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:56.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:03.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_12/index_7645_seed_19489_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:30:24.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 19:41:16.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16831_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 19:42:39.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:39.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:43:50.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_12/index_1340_seed_16832_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:44:15.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 19:55:08.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6182_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 19:56:29.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:29.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:57:41.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_12/index_219_seed_6183_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:58:03.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 20:09:02.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29913_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 20:10:23.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:23.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:35.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_12/index_6966_seed_29914_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:12:00.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.901 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 20:23:32.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:40.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27367_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 20:24:55.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:55.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:05.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_12/index_6514_seed_27368_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:26:26.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 20:37:18.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:19.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8791_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 20:38:43.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:43.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:39:48.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_12/index_3770_seed_8792_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:40:09.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 20:51:03.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7436_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 20:52:22.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:22.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:35.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_12/index_5367_seed_7437_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:53:57.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 21:04:50.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29113_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 21:06:14.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:06:14.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:23.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_12/index_7358_seed_29114_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:07:49.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 21:18:43.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:49.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1987_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 21:20:12.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:20:12.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:17.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_12/index_9321_seed_1988_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:21:39.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 21:32:31.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:37.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32332_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 21:33:52.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:52.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:01.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_12/index_2011_seed_32333_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:35:21.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 21:46:16.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:22.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20821_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 21:47:36.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:36.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:48:47.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_12/index_8922_seed_20822_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:49:10.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 22:00:05.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9955_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 22:01:25.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:25.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:37.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_12/index_2089_seed_9956_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:02:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 22:13:50.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6673_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 22:15:10.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:15:10.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:21.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_12/index_1156_seed_6674_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:16:45.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 22:27:39.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:45.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7233_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 22:29:05.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:29:05.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:10.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_12/index_634_seed_7234_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:30:32.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:58.080 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 22:41:59.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:41:59.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18723_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 22:43:22.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:22.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:31.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_12/index_1891_seed_18724_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:44:52.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 22:55:45.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:51.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9950_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 22:57:04.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:57:04.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_12/index_3731_seed_9951_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:58:39.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 23:09:30.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21359_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 23:10:52.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:52.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:01.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_12/index_3638_seed_21360_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:12:22.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 23:23:15.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16560_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 23:24:34.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:34.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:25:47.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_12/index_8787_seed_16561_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:26:11.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 23:37:07.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_27_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 23:38:28.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:28.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:39:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_12/index_4747_seed_28_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:40:02.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-25 23:51:52.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15324_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-25 23:53:17.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:53:17.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:23.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_12/index_4417_seed_15325_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:54:43.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 00:05:36.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9663_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 00:06:58.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:58.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:07.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_12/index_6432_seed_9664_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:08:28.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 00:19:21.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20499_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 00:20:45.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:45.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:21:50.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_12/index_3791_seed_20500_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:22:11.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 00:33:07.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:13.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14887_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 00:34:28.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:28.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:35:38.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_12/index_9885_seed_14888_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:36:04.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 00:47:03.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:09.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28205_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 00:48:25.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:25.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:37.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_12/index_8400_seed_28206_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:50:00.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:29.082 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 01:01:30.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:38.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17148_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 01:02:58.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:58.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:04.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_12/index_9552_seed_17149_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:04:25.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 01:15:17.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:17.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15659_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 01:16:38.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:38.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:17:50.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_12/index_5867_seed_15660_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:18:12.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 01:29:08.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:14.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10429_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 01:30:29.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:29.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:31:40.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_12/index_6589_seed_10430_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:32:03.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 01:42:58.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_1999_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 01:44:16.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:16.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:29.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_12/index_8642_seed_2000_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:45:50.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 01:56:44.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:50.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3068_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 01:58:05.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:58:05.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:17.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_12/index_3907_seed_3069_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:59:40.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.211 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 02:10:31.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:37.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12882_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 02:11:52.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:52.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:03.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_12/index_2912_seed_12883_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:13:25.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 02:24:16.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12611_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 02:25:40.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:40.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:26:46.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_12/index_4179_seed_12612_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:27:11.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 02:38:08.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8703_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 02:39:26.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:26.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:37.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_12/index_5470_seed_8704_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:41:00.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 02:51:53.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2227_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 02:53:15.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:53:15.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:25.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_12/index_271_seed_2228_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:54:48.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 03:05:40.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6627_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 03:07:02.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:07:02.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:14.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_12/index_6876_seed_6628_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:08:38.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.592 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 03:20:06.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17612_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 03:21:29.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:29.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:39.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_12/index_6942_seed_17613_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:22:59.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 03:33:52.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21422_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 03:35:14.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:35:14.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:23.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_12/index_3408_seed_21423_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:36:46.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 03:47:40.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26057_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 03:49:02.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:49:02.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:10.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_12/index_5120_seed_26058_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:50:31.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 04:01:21.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15760_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 04:02:43.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:43.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:03:51.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_12/index_6485_seed_15761_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:04:12.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 04:15:09.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18245_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 04:16:32.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:32.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:17:42.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_12/index_6658_seed_18246_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:18:05.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 04:29:03.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25139_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 04:30:28.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:28.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:33.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_12/index_9704_seed_25140_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:31:54.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 04:42:46.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8944_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 04:44:07.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:44:07.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:16.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_12/index_482_seed_8945_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:45:35.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 04:56:27.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21476_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 04:57:50.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:50.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:58:58.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_12/index_9076_seed_21477_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:59:19.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 05:10:10.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_97_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 05:11:30.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:30.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:12:41.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_12/index_6204_seed_98_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:13:03.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 05:24:03.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:09.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7488_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 05:25:28.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:28.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:35.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_12/index_5516_seed_7489_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:26:58.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.136 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 05:39:29.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5951_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 05:40:54.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:54.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:00.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_12/index_8904_seed_5952_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:42:20.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 05:53:10.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:16.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18298_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 05:54:32.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:32.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_12/index_1362_seed_18299_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:56:00.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 06:06:52.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:58.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18668_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 06:08:14.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:08:14.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:22.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_12/index_6908_seed_18669_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:09:44.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 06:20:40.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17710_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 06:22:01.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:22:01.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:09.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_12/index_1427_seed_17711_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:23:31.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 06:34:21.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25592_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 06:35:42.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:42.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:36:51.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_12/index_5241_seed_25593_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:37:11.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 06:48:09.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14040_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 06:49:27.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:27.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:50:44.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_12/index_21_seed_14041_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:51:03.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 07:02:03.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6820_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 07:03:24.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:24.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:33.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_12/index_5874_seed_6821_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:04:55.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 07:15:48.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:54.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14362_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 07:17:14.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:17:14.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:21.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_12/index_2840_seed_14363_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:18:41.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 07:29:32.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11085_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 07:30:53.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:53.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:04.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_12/index_9331_seed_11086_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:32:31.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 07:43:21.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9324_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 07:44:44.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:44.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:45:53.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_12/index_7671_seed_9325_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:46:15.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.709 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 07:57:45.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:53.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24003_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 07:59:08.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:59:08.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:18.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_12/index_6324_seed_24004_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:00:40.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 08:11:33.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:39.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7438_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 08:12:53.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:53.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:02.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_12/index_8605_seed_7439_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:14:26.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 08:25:18.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:24.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_960_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 08:26:41.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:41.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:27:47.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_12/index_1092_seed_961_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:28:08.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 08:39:03.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:09.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11097_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 08:40:28.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:28.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:35.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_12/index_1910_seed_11098_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:41:57.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 08:52:50.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13339_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 08:54:13.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:54:13.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_12/index_7091_seed_13340_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:55:45.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 09:06:39.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24634_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 09:08:03.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:08:03.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:11.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_12/index_904_seed_24635_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:09:33.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 09:20:22.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:23.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:28.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31475_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 09:21:46.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:46.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:22:52.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_12/index_9072_seed_31476_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:23:12.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 09:34:09.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17026_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 09:35:28.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:28.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:36:41.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_12/index_6217_seed_17027_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:37:04.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 09:47:56.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6381_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 09:49:16.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:16.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:29.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_12/index_8342_seed_6382_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:50:51.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 10:01:45.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:51.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33216_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 10:03:06.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:03:06.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:15.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_12/index_9042_seed_33217_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:04:35.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:16:00.116 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 10:16:01.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:09.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16435_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 10:17:21.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:21.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:34.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_12/index_642_seed_16436_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:18:59.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.655 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 10:29:52.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:58.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14142_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 10:31:11.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:31:11.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:23.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_12/index_6638_seed_14143_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:32:44.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 10:43:37.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1068_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 10:44:58.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:58.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:08.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_12/index_8412_seed_1069_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:46:29.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 10:57:21.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16677_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 10:58:42.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:42.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:59:56.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_12/index_9705_seed_16678_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:00:18.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 11:11:08.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3083_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 11:12:30.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:30.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:39.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_12/index_8237_seed_3084_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:13:59.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 11:26:08.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:14.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14049_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 11:27:28.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:28.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:28:40.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_12/index_7073_seed_14050_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:29:03.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 11:39:55.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10071_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 11:41:19.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:19.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:24.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_12/index_6717_seed_10072_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:42:44.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 11:53:37.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21021_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 11:54:59.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:59.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:07.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_12/index_8978_seed_21022_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:56:31.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 12:07:21.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8838_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 12:08:43.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:43.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:09:51.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_12/index_7386_seed_8839_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:10:12.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 12:21:06.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14210_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 12:22:26.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:38.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_12/index_8500_seed_14211_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:24:00.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:29.025 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 12:35:30.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:38.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32870_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 12:36:53.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:53.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:03.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_12/index_2895_seed_32871_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:38:25.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 12:49:16.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7824_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 12:50:38.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:38.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:51:48.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_12/index_4800_seed_7825_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:52:09.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 13:03:03.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5779_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 13:04:26.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:26.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:33.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_12/index_8104_seed_5780_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:05:53.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 13:16:44.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11198_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 13:18:08.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:18:08.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:16.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_12/index_5190_seed_11199_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:19:41.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 13:30:31.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:37.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20780_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 13:31:56.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:56.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:04.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_12/index_4427_seed_20781_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:33:25.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 13:44:15.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:21.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29859_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 13:45:35.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:35.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:46:49.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_12/index_4911_seed_29860_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:47:13.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 13:58:07.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:13.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20793_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 13:59:26.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:26.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:36.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_12/index_3934_seed_20794_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:00:57.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 14:11:49.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:55.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12960_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 14:13:10.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:13:10.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:18.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_12/index_4705_seed_12961_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:14:39.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 14:25:28.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32888_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 14:26:46.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:46.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:02.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_12/index_1249_seed_32889_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:28:25.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 14:39:15.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9543_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 14:40:35.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:35.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:41:50.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_12/index_8425_seed_9544_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:42:13.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.535 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 14:53:45.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11842_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 14:55:06.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:55:06.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:16.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_12/index_1727_seed_11843_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:56:36.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 15:07:27.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:33.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19513_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 15:08:46.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:46.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:09:56.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_12/index_1126_seed_19514_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:10:19.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 15:21:11.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29115_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 15:22:29.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:29.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:23:41.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_12/index_1151_seed_29116_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:24:02.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 15:34:56.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18915_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 15:36:16.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:16.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:29.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_12/index_5266_seed_18916_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:37:50.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 15:48:45.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:51.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24558_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 15:50:05.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:50:05.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:15.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_12/index_1846_seed_24559_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:51:35.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 16:02:26.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12257_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 16:03:47.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:47.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:04:57.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_12/index_1273_seed_12258_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:05:17.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.002 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 16:16:06.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18521_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 16:17:29.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:29.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:18:42.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_12/index_5924_seed_18522_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:19:04.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 16:29:54.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:30:59.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15433_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 16:31:14.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:14.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:24.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_12/index_840_seed_15434_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:32:45.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 16:43:38.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:44.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16701_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 16:44:57.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:57.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:08.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_12/index_3163_seed_16702_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:46:31.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 16:57:20.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:21.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:25.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25197_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 16:58:43.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:43.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:59:54.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_12/index_9555_seed_25198_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:00:15.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.086 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 17:12:53.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10951_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 17:14:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:14:15.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:25.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_12/index_6523_seed_10952_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:15:48.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 17:26:39.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_12/index_1054_seed_32583_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 17:28:01.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:28:01.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:09.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_12/index_1054_seed_32584_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:29:30.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 17:40:20.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_12/index_9232_seed_25733_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 17:41:48.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:48.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:42:54.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_12/index_9232_seed_25734_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:43:14.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 17:54:06.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:12.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_12/index_9258_seed_5216_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 17:55:27.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:27.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:37.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_12/index_9258_seed_5217_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:56:58.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 18:07:49.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_12/index_5642_seed_7592_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 18:09:09.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:09:09.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:21.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_12/index_5642_seed_7593_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:10:46.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 18:21:38.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:44.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_12/index_537_seed_12605_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 18:22:57.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:57.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:12.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_12/index_537_seed_12606_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:24:37.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.994 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 18:35:27.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_12/index_3311_seed_18938_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 18:36:48.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:48.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:03.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_12/index_3311_seed_18939_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:38:24.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 18:49:15.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_12/index_1849_seed_1099_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 18:50:34.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:34.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:51:45.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_12/index_1849_seed_1100_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:52:05.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 19:03:01.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:01.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_12/index_4467_seed_17953_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 19:04:24.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:24.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:35.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_12/index_4467_seed_17954_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:05:55.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 19:16:46.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_12/index_5836_seed_7703_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 19:18:11.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:18:11.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:19.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_12/index_5836_seed_7704_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:19:40.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.713 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 19:31:08.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_12/index_6728_seed_3948_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 19:32:32.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:32.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:33:43.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_12/index_6728_seed_3949_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:34:07.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 19:45:03.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:09.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_12/index_9620_seed_15361_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 19:46:24.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:24.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:32.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_12/index_9620_seed_15362_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:47:52.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 19:58:45.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_12/index_9972_seed_4736_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 20:00:06.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:00:06.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:17.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_12/index_9972_seed_4737_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:01:37.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 20:12:28.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_12/index_6474_seed_23920_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 20:13:50.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:50.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:14:58.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_12/index_6474_seed_23921_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:15:22.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 20:26:14.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_12/index_5337_seed_31080_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 20:27:35.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:35.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:28:48.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_12/index_5337_seed_31081_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:29:12.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 20:40:07.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_12/index_9961_seed_11396_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 20:41:26.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:26.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:42:38.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_12/index_9961_seed_11397_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:43:01.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 20:53:52.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_12/index_5940_seed_12277_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 20:55:13.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:55:13.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:25.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_12/index_5940_seed_12278_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:56:54.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 21:07:44.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_12/index_2476_seed_19990_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 21:09:05.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:09:05.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:18.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_12/index_2476_seed_19991_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:10:42.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 21:21:31.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_12/index_8267_seed_16725_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 21:22:51.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:51.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:02.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_12/index_8267_seed_16726_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:24:26.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 21:35:16.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:16.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_12/index_2563_seed_18235_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 21:36:38.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:38.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:37:49.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_12/index_2563_seed_18236_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:38:11.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.575 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 21:49:43.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:43.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:51.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_12/index_3464_seed_3232_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 21:51:05.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:51:05.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:15.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_12/index_3464_seed_3233_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:52:38.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 22:03:26.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:27.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_12/index_3563_seed_7371_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 22:04:48.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:48.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:05:55.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_12/index_3563_seed_7372_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:06:18.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 22:17:07.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_12/index_2956_seed_20297_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 22:18:29.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:29.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:36.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_12/index_2956_seed_20298_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:19:59.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 22:30:49.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_12/index_288_seed_17072_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 22:32:11.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:32:11.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:20.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_12/index_288_seed_17073_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:33:43.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 22:44:32.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_12/index_677_seed_16517_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 22:45:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:54.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:03.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_12/index_677_seed_16518_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:47:24.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 22:59:16.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_12/index_1914_seed_31885_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 23:00:35.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:35.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:01:48.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_12/index_1914_seed_31886_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:02:09.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 23:13:02.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_12/index_707_seed_508_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 23:14:23.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:23.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:31.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_12/index_707_seed_509_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:15:53.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 23:26:44.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_12/index_2627_seed_16084_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 23:28:03.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:28:03.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:15.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_12/index_2627_seed_16085_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:29:39.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 23:40:27.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:33.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_12/index_292_seed_32695_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 23:41:45.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:45.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:01.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_12/index_292_seed_32696_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:43:26.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.523 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-26 23:54:16.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_12/index_5639_seed_4507_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-26 23:55:36.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:36.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:56:46.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_12/index_5639_seed_4508_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:57:09.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.433 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 00:08:40.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:48.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_12/index_3641_seed_32927_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 00:10:05.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:10:05.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:11.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_12/index_3641_seed_32928_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:11:32.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 00:22:20.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_12/index_6526_seed_17398_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 00:23:39.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:39.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:24:53.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_12/index_6526_seed_17399_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 00:25:14.490 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:25:15.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 00:36:06.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_12/index_1845_seed_12329_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 00:37:28.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:28.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:38.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_12/index_1845_seed_12330_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:38:59.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 00:49:49.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_12/index_7487_seed_3199_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 00:51:13.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:51:13.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:19.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_12/index_7487_seed_3200_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 00:52:43.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:52:44.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 01:03:35.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_12/index_6071_seed_8813_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 01:04:57.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:57.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:05.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_12/index_6071_seed_8814_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:06:28.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 01:17:19.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_12/index_7984_seed_21254_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 01:18:40.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:40.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:19:52.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_12/index_7984_seed_21255_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:20:15.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 01:31:07.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_12/index_1104_seed_10403_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 01:32:28.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:28.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:38.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_12/index_1104_seed_10404_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:33:58.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 01:44:47.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:52.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_12/index_8052_seed_28648_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 01:46:10.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:46:10.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:19.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_12/index_8052_seed_28649_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:47:40.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 01:58:27.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_12/index_6867_seed_7133_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 01:59:48.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:48.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:01.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_12/index_6867_seed_7134_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:01:23.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 02:12:14.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_12/index_4498_seed_11025_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 02:13:33.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:33.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:14:47.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_12/index_4498_seed_11026_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:15:09.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.196 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 02:26:39.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_12/index_6746_seed_5317_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 02:28:00.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:28:00.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:12.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_12/index_6746_seed_5318_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:29:34.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 02:40:23.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_12/index_3944_seed_30078_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 02:41:43.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:43.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:00.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_12/index_3944_seed_30079_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:43:25.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 02:54:14.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:20.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_12/index_2489_seed_5420_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 02:55:40.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:40.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:56:46.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_12/index_2489_seed_5421_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:57:09.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 03:08:01.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:02.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_12/index_1887_seed_9313_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 03:09:21.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:21.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:31.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_12/index_1887_seed_9314_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:10:50.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 03:21:41.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:48.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_12/index_1720_seed_32326_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 03:23:03.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:23:03.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:15.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_12/index_1720_seed_32327_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:24:38.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 03:35:26.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_12/index_5613_seed_15018_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 03:36:48.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:48.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:02.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_12/index_5613_seed_15019_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:38:25.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 03:49:14.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:20.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_12/index_3067_seed_23564_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 03:50:34.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:34.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:51:45.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_12/index_3067_seed_23565_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:52:08.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 04:03:03.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_12/index_5391_seed_25676_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 04:04:22.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:22.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:33.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_12/index_5391_seed_25677_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:05:54.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 04:16:48.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_12/index_8189_seed_24536_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 04:18:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:18:06.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:20.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_12/index_8189_seed_24537_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:19:42.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 04:30:31.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_12/index_8779_seed_13713_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 04:31:52.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:52.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:01.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_12/index_8779_seed_13714_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:33:22.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.569 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.288 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 04:45:58.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_12/index_6441_seed_1668_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 04:47:20.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:20.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:31.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_12/index_6441_seed_1669_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:48:56.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 04:59:46.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_12/index_5224_seed_7991_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 05:01:07.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:01:07.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:17.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_12/index_5224_seed_7992_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:02:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 05:13:28.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:33.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_12/index_1605_seed_9325_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 05:14:48.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:48.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:15:56.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_12/index_1605_seed_9326_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:16:17.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 05:27:07.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:12.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_12/index_7303_seed_6557_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 05:28:27.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:27.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:29:39.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_12/index_7303_seed_6558_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:30:01.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 05:40:50.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_12/index_4791_seed_22749_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 05:42:10.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:42:10.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:19.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_12/index_4791_seed_22750_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:43:44.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 05:54:35.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_12/index_999_seed_32065_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 05:55:55.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:55.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:06.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_12/index_999_seed_32066_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:57:30.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 06:08:19.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_12/index_54_seed_17048_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 06:09:39.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:39.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:10:50.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_12/index_54_seed_17049_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:11:14.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.063 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 06:22:07.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:12.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_12/index_3738_seed_8867_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 06:23:28.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:28.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:38.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_12/index_3738_seed_8868_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:24:59.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.196 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 06:35:50.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_12/index_9420_seed_19563_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 06:37:11.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:37:11.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:20.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_12/index_9420_seed_19564_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:38:42.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 06:49:30.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_12/index_1078_seed_21832_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 06:50:50.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:50.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_12/index_1078_seed_21833_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:52:21.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:12.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.704 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 07:03:49.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_12/index_6744_seed_30156_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 07:05:14.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:05:14.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:20.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_12/index_6744_seed_30157_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:06:40.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 07:17:29.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_12/index_2138_seed_6828_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 07:18:49.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:49.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:00.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_12/index_2138_seed_6829_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:20:22.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 07:31:16.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_12/index_6274_seed_20849_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 07:32:35.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:35.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:33:49.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_12/index_6274_seed_20850_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:34:12.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 07:45:07.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_12/index_8130_seed_16723_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 07:46:30.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:30.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:47:38.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_12/index_8130_seed_16724_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 07:48:01.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:48:02.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 07:58:52.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_12/index_8050_seed_1421_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 08:00:16.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:00:16.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:23.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_12/index_8050_seed_1422_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:01:46.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 08:12:38.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_12/index_5871_seed_5583_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 08:14:00.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:14:00.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:10.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_12/index_5871_seed_5584_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:15:31.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 08:26:20.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_12/index_541_seed_6567_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 08:27:42.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:42.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:28:51.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_12/index_541_seed_6568_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:29:11.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 08:40:03.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_12/index_1119_seed_30142_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 08:41:21.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:21.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:35.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_12/index_1119_seed_30143_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:42:57.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.092 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 08:53:47.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_12/index_5920_seed_26294_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 08:55:06.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:55:06.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_12/index_5920_seed_26295_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:56:50.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 09:07:41.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:48.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_12/index_8993_seed_26287_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 09:09:02.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:09:02.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:13.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_12/index_8993_seed_26288_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:10:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.071 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 09:22:02.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:02.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_12/index_3591_seed_18992_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 09:23:25.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:25.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:34.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_12/index_3591_seed_18993_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:24:55.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 09:35:43.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:49.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_12/index_7061_seed_29947_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 09:37:03.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:37:03.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:13.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_12/index_7061_seed_29948_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:38:35.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 09:49:26.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_12/index_9832_seed_30162_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 09:50:49.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:49.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:51:57.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_12/index_9832_seed_30163_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:52:17.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 10:03:08.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_12/index_6602_seed_21106_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 10:04:25.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:25.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:39.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_12/index_6602_seed_21107_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:06:00.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 10:16:53.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_12/index_9366_seed_1672_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 10:18:13.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:18:13.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:24.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_12/index_9366_seed_1673_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:19:46.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 10:31:49.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_12/index_1379_seed_7597_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 10:33:09.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:33:09.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:21.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_12/index_1379_seed_7598_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:34:43.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 10:45:31.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_12/index_9950_seed_4815_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 10:46:55.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:55.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:03.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_12/index_9950_seed_4816_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:48:27.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 10:59:16.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_12/index_7549_seed_11511_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 11:00:34.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:34.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:01:49.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_12/index_7549_seed_11512_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:02:13.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 11:13:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_12/index_5810_seed_20412_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 11:14:27.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:27.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:35.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_12/index_5810_seed_20413_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:15:55.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 11:26:47.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_12/index_8781_seed_30535_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 11:28:09.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:28:09.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:21.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_12/index_8781_seed_30536_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:29:42.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.404 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 11:41:06.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_12/index_4065_seed_25996_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 11:42:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:29.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:37.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_12/index_4065_seed_25997_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:43:59.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 11:54:48.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_12/index_7359_seed_26217_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 11:56:07.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:56:07.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:19.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_12/index_7359_seed_26218_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:57:45.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 12:08:34.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_12/index_6393_seed_10484_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 12:10:00.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:10:00.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:05.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_12/index_6393_seed_10485_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:11:27.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.655 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 12:22:19.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_12/index_7540_seed_28065_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 12:23:44.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:44.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:24:50.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_12/index_7540_seed_28066_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:25:10.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 12:36:01.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_12/index_4508_seed_19225_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 12:37:21.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:21.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:32.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_12/index_4508_seed_19226_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:38:53.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 12:49:44.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:50.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_12/index_9662_seed_26239_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 12:51:03.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:51:03.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:14.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_12/index_9662_seed_26240_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:52:34.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 13:03:21.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_12/index_5494_seed_7712_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 13:04:42.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:42.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:05:51.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_12/index_5494_seed_7713_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:06:13.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 13:17:07.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_12/index_3800_seed_25180_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 13:18:28.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:28.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:37.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_12/index_3800_seed_25181_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:19:58.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 13:30:47.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_12/index_5161_seed_22807_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 13:32:07.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:32:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:21.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_12/index_5161_seed_22808_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:33:47.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 13:44:38.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_12/index_9444_seed_8914_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 13:45:59.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:59.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:09.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_12/index_9444_seed_8915_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:47:31.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.220 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.271 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 13:58:57.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_12/index_7839_seed_29017_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 14:00:16.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:16.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:30.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_12/index_7839_seed_29018_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:01:51.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 14:12:43.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:49.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_12/index_1589_seed_28711_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 14:14:04.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:14:04.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:12.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_12/index_1589_seed_28712_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:15:32.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 14:26:20.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_12/index_7641_seed_23056_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 14:27:39.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:39.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:28:50.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_12/index_7641_seed_23057_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:29:12.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 14:40:05.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_12/index_253_seed_10527_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 14:41:29.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:29.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:36.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_12/index_253_seed_10528_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:42:58.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 14:53:47.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_12/index_5248_seed_16275_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 14:55:12.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:55:12.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:18.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_12/index_5248_seed_16276_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:56:39.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.811 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 15:07:26.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:26.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:31.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_12/index_2659_seed_23577_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 15:08:51.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:51.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:09:56.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_12/index_2659_seed_23578_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:10:18.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 15:21:06.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:12.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_12/index_1796_seed_4889_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 15:22:26.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:26.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:37.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_12/index_1796_seed_4890_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:23:58.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 15:34:47.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_12/index_6772_seed_30407_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 15:36:09.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:36:09.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:20.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_12/index_6772_seed_30408_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:37:44.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 15:48:32.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_12/index_3240_seed_22228_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 15:49:51.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:51.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:03.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_12/index_3240_seed_22229_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:51:25.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 16:02:14.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_12/index_3479_seed_1530_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 16:03:33.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:33.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:04:44.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_12/index_3479_seed_1531_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:05:05.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.764 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 16:17:43.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_12/index_6301_seed_19603_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 16:19:07.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:19:07.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:17.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_12/index_6301_seed_19604_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:20:37.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 16:31:26.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_12/index_9466_seed_8663_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 16:32:53.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:53.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:33:58.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_12/index_9466_seed_8664_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:34:18.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 16:45:07.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_12/index_9391_seed_9238_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 16:46:28.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:28.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:38.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_12/index_9391_seed_9239_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:47:59.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 16:58:47.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:53.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_12/index_92_seed_28632_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 17:00:08.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:00:08.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:20.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_12/index_92_seed_28633_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:01:42.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 17:12:31.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_12/index_1835_seed_27716_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 17:13:51.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:51.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:02.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_12/index_1835_seed_27717_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:15:22.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 17:26:15.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_12/index_8640_seed_33521_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 17:27:35.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:35.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:28:46.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_12/index_8640_seed_33522_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:29:09.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 17:40:00.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:00.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:06.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_12/index_7740_seed_4044_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 17:41:22.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:22.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:34.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_12/index_7740_seed_4045_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:42:55.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 17:53:45.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:51.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_12/index_2878_seed_11193_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 17:55:05.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:55:05.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:15.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_12/index_2878_seed_11194_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:56:38.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 18:07:25.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_12/index_4318_seed_29601_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 18:08:51.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:51.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:01.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_12/index_4318_seed_29602_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:10:24.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 18:21:14.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_12/index_2788_seed_4404_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 18:22:39.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:39.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:23:45.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_12/index_2788_seed_4405_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:24:07.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.432 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 18:35:38.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:44.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_12/index_3280_seed_6313_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 18:36:59.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:59.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:09.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_12/index_3280_seed_6314_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:38:32.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 18:49:21.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_12/index_9460_seed_30765_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 18:50:41.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:41.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:51:55.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_12/index_9460_seed_30766_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:52:16.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 19:03:06.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:12.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_12/index_4832_seed_22487_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 19:04:25.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:25.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:36.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_12/index_4832_seed_22488_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:05:55.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 19:16:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_12/index_2067_seed_29201_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 19:18:04.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:18:04.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:21.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_12/index_2067_seed_29202_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:19:42.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 19:30:30.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_12/index_6156_seed_30086_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 19:31:50.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:50.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:03.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_12/index_6156_seed_30087_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:33:25.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 19:44:14.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_12/index_997_seed_21138_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 19:45:32.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:32.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:46:44.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_12/index_997_seed_21139_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:47:10.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 19:58:02.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:08.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_12/index_7526_seed_29817_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 19:59:25.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:25.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:33.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_12/index_7526_seed_29818_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:00:54.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 20:11:44.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_12/index_9553_seed_18418_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 20:13:04.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:13:04.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:15.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_12/index_9553_seed_18419_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:14:39.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 20:25:26.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_12/index_1783_seed_19070_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 20:26:49.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:49.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:27:58.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_12/index_1783_seed_19071_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:28:19.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 20:39:06.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:07.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_12/index_7902_seed_22619_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 20:40:27.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:27.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:41:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_12/index_7902_seed_22620_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:42:04.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.519 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 20:53:29.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:35.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_12/index_7699_seed_18110_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 20:54:51.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:51.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:55:59.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_12/index_7699_seed_18111_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:56:21.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 21:07:10.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:16.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_12/index_2123_seed_15495_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 21:08:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:29.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:09:41.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_12/index_2123_seed_15496_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:10:07.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 21:21:01.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_12/index_3616_seed_23085_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 21:22:22.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:22.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:31.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_12/index_3616_seed_23086_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:23:52.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 21:34:42.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:48.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_12/index_8502_seed_15658_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 21:36:05.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:36:05.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:12.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_12/index_8502_seed_15659_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:37:34.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 21:48:24.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:30.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_12/index_8175_seed_5846_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 21:49:46.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:46.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:50:54.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_12/index_8175_seed_5847_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 21:51:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:51:16.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 22:03:21.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_12/index_423_seed_26957_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 22:04:43.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:43.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:05:54.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_12/index_423_seed_26958_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:06:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 22:17:06.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:07.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:12.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_12/index_8675_seed_11562_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 22:18:25.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:25.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:36.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_12/index_8675_seed_11563_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 22:19:56.152 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:19:57.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 22:30:47.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_12/index_5989_seed_22421_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 22:32:07.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:32:07.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:21.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_12/index_5989_seed_22422_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:33:41.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 22:44:30.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_12/index_8379_seed_3875_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 22:45:52.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:52.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:46:59.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_12/index_8379_seed_3876_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:47:18.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 22:58:07.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_12/index_5783_seed_24221_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 22:59:25.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:25.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:00:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_12/index_5783_seed_24222_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:01:01.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.363 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 23:12:29.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_12/index_6212_seed_16385_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 23:13:51.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:51.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:01.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_12/index_6212_seed_16386_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:15:23.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 23:26:15.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_12/index_5436_seed_13662_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 23:27:34.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:34.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:28:45.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_12/index_5436_seed_13663_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:29:06.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 23:40:02.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_12/index_1085_seed_2395_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 23:41:23.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:23.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:42:36.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_12/index_1085_seed_2396_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:43:01.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-27 23:53:51.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_12/index_4182_seed_25127_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-27 23:55:11.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:55:11.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:22.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_12/index_4182_seed_25128_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:56:44.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 00:07:32.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:38.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_12/index_5170_seed_8515_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 00:08:53.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:53.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:08.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_12/index_5170_seed_8516_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:10:31.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 00:21:21.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_12/index_3215_seed_12330_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 00:22:42.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:42.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:23:51.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_12/index_3215_seed_12331_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:24:12.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 00:35:07.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_12/index_9269_seed_28390_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 00:36:29.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:29.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:37:45.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_12/index_9269_seed_28391_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:38:10.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 00:49:02.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:03.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:08.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_12/index_7053_seed_7162_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 00:50:22.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:22.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:36.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_12/index_7053_seed_7163_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:51:57.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 01:02:48.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:54.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_12/index_6098_seed_10675_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 01:04:09.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:04:09.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:21.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_12/index_6098_seed_10676_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:05:44.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 01:16:33.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_12/index_3813_seed_25324_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 01:17:57.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:11.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_12/index_3813_seed_25325_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:19:31.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.790 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 01:30:58.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_12/index_1011_seed_12180_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 01:32:22.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:22.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_12/index_1011_seed_12181_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:33:52.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 01:44:44.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_12/index_8597_seed_6388_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 01:46:04.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:46:04.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:16.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_12/index_8597_seed_6389_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:47:37.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 01:58:27.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_12/index_8225_seed_20150_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 01:59:47.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:47.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:00:58.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_12/index_8225_seed_20151_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:01:18.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 02:12:08.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_12/index_9653_seed_1591_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 02:13:32.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:14:40.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_12/index_9653_seed_1592_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:15:04.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 02:25:53.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_12/index_2327_seed_9319_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 02:27:15.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:27:15.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:25.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_12/index_2327_seed_9320_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:28:45.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 02:39:38.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:44.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_12/index_1079_seed_23178_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 02:40:59.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:59.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:08.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_12/index_1079_seed_23179_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:42:31.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 02:53:20.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:21.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_12/index_180_seed_15563_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 02:54:41.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:41.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:55:51.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_12/index_180_seed_15564_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:56:13.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 03:07:08.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_12/index_1094_seed_17886_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 03:08:31.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:31.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:09:38.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_12/index_1094_seed_17887_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:10:03.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.615 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 03:20:53.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:59.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_12/index_905_seed_14227_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 03:22:13.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:22:13.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:26.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_12/index_905_seed_14228_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:23:49.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 03:34:40.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_12/index_1122_seed_8561_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 03:36:01.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:36:01.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:12.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_12/index_1122_seed_8562_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:37:35.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.696 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 03:50:13.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_12/index_132_seed_6732_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 03:51:38.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:38.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:52:46.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_12/index_132_seed_6733_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:53:06.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 04:04:03.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_12/index_1207_seed_21581_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 04:05:26.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:26.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_12/index_1207_seed_21582_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:06:56.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 04:17:48.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_12/index_8382_seed_32195_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 04:19:08.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:19:08.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:20.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_12/index_8382_seed_32196_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:20:46.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.283 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 04:31:40.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_12/index_3423_seed_5097_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 04:32:59.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:59.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:09.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_12/index_3423_seed_5098_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:34:29.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 04:45:20.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_12/index_6043_seed_3447_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 04:46:39.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:39.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:47:50.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_12/index_6043_seed_3448_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:48:11.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 04:59:03.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_12/index_2589_seed_30835_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 05:00:23.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:23.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:32.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_12/index_2589_seed_30836_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:01:52.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 05:12:44.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:45.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_12/index_8043_seed_23228_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 05:14:05.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:14:05.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:14.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_12/index_8043_seed_23229_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:15:35.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 05:26:26.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_12/index_539_seed_22505_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 05:27:45.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:45.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:28:56.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_12/index_539_seed_22506_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:29:22.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 05:40:15.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_12/index_1823_seed_5348_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 05:41:38.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:38.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:42:47.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_12/index_1823_seed_5349_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:43:10.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 05:54:07.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:07.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_12/index_7541_seed_25259_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 05:55:32.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:32.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:38.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_12/index_7541_seed_25260_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:57:00.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.779 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 06:08:27.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:33.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_12/index_5645_seed_29954_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 06:09:47.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:47.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:10:58.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_12/index_5645_seed_29955_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:11:21.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 06:22:11.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_12/index_915_seed_26647_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 06:23:34.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:34.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:24:42.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_12/index_915_seed_26648_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:25:07.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 06:36:02.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_12/index_2774_seed_1807_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 06:37:22.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:22.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:32.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_12/index_2774_seed_1808_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:38:53.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 06:49:45.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_12/index_627_seed_26225_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 06:51:05.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:51:05.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:16.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_12/index_627_seed_26226_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:52:37.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 07:03:27.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:33.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_12/index_2470_seed_23681_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 07:04:50.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:50.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:05:59.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_12/index_2470_seed_23682_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:06:21.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 07:17:12.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_12/index_5332_seed_27777_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 07:18:31.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:31.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:19:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_12/index_5332_seed_27778_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:20:04.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 07:30:53.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:59.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_12/index_2581_seed_11489_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 07:32:16.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:32:16.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:26.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_12/index_2581_seed_11490_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:33:48.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 07:44:40.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:46.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_12/index_4185_seed_3331_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 07:46:06.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:46:06.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:11.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_12/index_4185_seed_3332_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:47:33.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 07:58:21.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:26.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_12/index_5911_seed_15193_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 07:59:44.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:44.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:00:52.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_12/index_5911_seed_15194_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:01:16.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 08:12:08.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:14.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_12/index_4586_seed_11077_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 08:13:29.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:29.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:14:39.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_12/index_4586_seed_11078_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:15:03.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.307 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 08:26:31.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:37.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_12/index_3785_seed_17033_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 08:27:52.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:02.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_12/index_3785_seed_17034_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:29:27.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 08:40:22.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_12/index_2790_seed_19717_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 08:41:46.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:46.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:42:53.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_12/index_2790_seed_19718_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:43:14.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 08:54:08.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_12/index_833_seed_29174_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 08:55:29.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:29.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:38.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_12/index_833_seed_29175_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:57:00.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 09:07:50.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:51.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_12/index_3968_seed_29_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 09:09:10.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:09:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:21.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_12/index_3968_seed_30_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:10:44.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 09:21:32.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:38.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_12/index_5599_seed_24574_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 09:22:55.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:55.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:04.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_12/index_5599_seed_24575_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:24:28.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 09:36:35.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_12/index_669_seed_17783_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 09:37:56.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:56.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:04.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_12/index_669_seed_17784_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 09:39:26.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:39:27.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 09:50:22.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_12/index_4942_seed_11390_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 09:51:42.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:42.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:52:55.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_12/index_4942_seed_11391_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:53:25.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 10:04:16.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:22.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_12/index_3011_seed_1201_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 10:05:45.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:45.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:06:50.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_12/index_3011_seed_1202_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:07:11.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 10:18:09.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_12/index_7613_seed_1169_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 10:19:29.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:29.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:20:40.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_12/index_7613_seed_1170_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:21:02.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.312 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 10:31:54.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_12/index_6374_seed_16297_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 10:33:14.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:33:14.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:26.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_12/index_6374_seed_16298_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:34:54.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.824 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 10:46:20.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_12/index_137_seed_11717_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 10:47:43.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:43.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:48:51.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_12/index_137_seed_11718_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:49:13.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 11:00:08.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_12/index_424_seed_26194_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 11:01:27.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:27.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:39.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_12/index_424_seed_26195_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:03:00.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 11:13:51.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_12/index_2754_seed_3286_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 11:15:11.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:15:11.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:21.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_12/index_2754_seed_3287_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:16:43.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 11:27:32.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:38.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_12/index_1318_seed_20780_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 11:28:53.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:53.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:02.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_12/index_1318_seed_20781_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:30:23.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 11:41:15.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_12/index_7094_seed_16976_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 11:42:34.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:34.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:43:47.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_12/index_7094_seed_16977_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:44:08.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 11:55:03.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:09.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_12/index_3666_seed_23409_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 11:56:28.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:28.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:33.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_12/index_3666_seed_23410_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:57:57.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 12:08:48.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:54.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_12/index_6765_seed_30386_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 12:10:10.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:10:10.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:25.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_12/index_6765_seed_30387_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:11:48.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 12:22:40.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:46.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_12/index_1258_seed_18359_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 12:24:01.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:24:01.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:13.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_12/index_1258_seed_18360_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:25:34.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 12:36:24.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:30.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_12/index_8030_seed_28556_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 12:37:43.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:43.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:38:55.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_12/index_8030_seed_28557_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:39:17.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 12:50:08.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:09.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_12/index_8943_seed_15116_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 12:51:28.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:28.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:52:41.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_12/index_8943_seed_15117_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:53:04.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.695 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 13:04:33.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_12/index_7725_seed_25574_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 13:05:55.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:55.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:07.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_12/index_7725_seed_25575_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 13:07:33.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:07:33.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 13:18:22.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_12/index_2036_seed_8002_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 13:19:45.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:45.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:20:53.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_12/index_2036_seed_8003_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:21:14.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 13:32:07.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_12/index_4757_seed_14437_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 13:33:26.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:26.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:38.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_12/index_4757_seed_14438_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:35:00.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 13:45:49.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_12/index_4306_seed_5850_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 13:47:09.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:47:09.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:22.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_12/index_4306_seed_5851_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:48:42.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 13:59:31.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_12/index_4211_seed_30517_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 14:00:49.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:49.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:00.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_12/index_4211_seed_30518_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:02:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 14:13:08.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:14.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_12/index_9012_seed_9745_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 14:14:29.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:29.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:15:42.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_12/index_9012_seed_9746_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:16:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 14:27:02.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_12/index_4672_seed_7618_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 14:28:23.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:23.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:31.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_12/index_4672_seed_7619_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:29:51.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 14:40:44.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_12/index_6130_seed_4673_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 14:42:07.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:42:07.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:15.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_12/index_6130_seed_4674_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:43:36.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 14:54:25.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:31.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_12/index_8010_seed_29108_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 14:55:46.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:46.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:56:55.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_12/index_8010_seed_29109_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:57:15.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 15:08:07.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_12/index_6799_seed_26909_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 15:09:26.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:26.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:10:39.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_12/index_6799_seed_26910_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:11:02.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.186 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 15:23:36.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_12/index_8133_seed_24972_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 15:24:56.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:56.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:08.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_12/index_8133_seed_24973_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:26:32.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 15:37:21.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_12/index_7242_seed_28363_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 15:38:45.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:45.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:39:52.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_12/index_7242_seed_28364_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:40:14.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 15:51:08.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_12/index_1479_seed_32933_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 15:52:27.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:27.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:53:40.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_12/index_1479_seed_32934_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:54:04.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 16:04:54.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_12/index_2707_seed_12810_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 16:06:14.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:06:14.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:24.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_12/index_2707_seed_12811_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:07:50.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.994 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 16:18:40.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:41.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_12/index_5251_seed_31159_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 16:20:03.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:20:03.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:12.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_12/index_5251_seed_31160_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:21:34.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 16:32:23.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_12/index_9755_seed_4685_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 16:33:43.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:43.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:34:53.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_12/index_9755_seed_4686_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:35:14.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 16:46:07.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:13.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_12/index_4876_seed_14301_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 16:47:26.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:26.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:39.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_12/index_4876_seed_14302_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:49:00.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 16:59:50.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_12/index_790_seed_19723_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 17:01:11.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:01:11.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:21.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_12/index_790_seed_19724_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 17:02:43.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:02:44.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 17:13:34.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:40.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_12/index_2092_seed_27850_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 17:14:56.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:56.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:08.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_12/index_2092_seed_27851_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:16:31.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 17:27:21.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_12/index_9799_seed_6857_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 17:28:43.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:43.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:29:53.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_12/index_9799_seed_6858_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:30:18.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.805 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 17:41:44.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_12/index_3506_seed_27582_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 17:43:03.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:43:03.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:20.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_12/index_3506_seed_27583_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:44:43.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 17:55:31.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_12/index_8798_seed_30336_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 17:56:50.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:50.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:02.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_12/index_8798_seed_30337_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:58:25.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 18:09:15.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_12/index_3000_seed_18305_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 18:10:35.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:35.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:11:49.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_12/index_3000_seed_18306_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:12:10.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 18:23:03.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_12/index_2640_seed_1136_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 18:24:25.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:25.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:33.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_12/index_2640_seed_1137_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:25:54.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.092 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 18:36:44.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_12/index_646_seed_28684_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 18:38:06.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:38:06.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:16.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_12/index_646_seed_28685_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:39:39.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 18:50:26.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_12/index_4814_seed_6750_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 18:51:45.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:45.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:52:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_12/index_4814_seed_6751_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:53:22.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 19:04:13.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_12/index_9792_seed_33193_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 19:05:35.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:35.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:06:45.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_12/index_9792_seed_33194_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:07:08.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 19:18:02.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_12/index_5771_seed_18635_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 19:19:21.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:21.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:30.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_12/index_5771_seed_18636_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:20:53.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 19:31:44.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:50.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_12/index_5024_seed_14615_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 19:33:05.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:33:05.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:16.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_12/index_5024_seed_14616_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 19:34:36.662 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:34:37.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 19:45:24.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_12/index_4444_seed_32511_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 19:46:45.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:45.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:47:56.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_12/index_4444_seed_32512_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:48:18.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.706 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 19:59:44.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_12/index_44_seed_19113_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 20:01:02.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:01:02.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:15.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_12/index_44_seed_19114_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:02:35.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 20:13:25.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_12/index_4387_seed_23847_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 20:14:45.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:45.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:15:55.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_12/index_4387_seed_23848_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:16:16.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 20:27:07.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_12/index_9474_seed_21673_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 20:28:33.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:33.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:38.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_12/index_9474_seed_21674_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:29:59.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 20:40:49.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_12/index_8891_seed_4677_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 20:42:07.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:42:07.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:22.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_12/index_8891_seed_4678_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:43:46.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 20:54:36.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_12/index_6905_seed_21523_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 20:55:55.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:55.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:09.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_12/index_6905_seed_21524_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:57:34.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 21:09:41.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_12/index_7353_seed_12555_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 21:11:02.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:11:02.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:14.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_12/index_7353_seed_12556_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:12:35.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 21:23:24.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_12/index_4650_seed_22607_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 21:24:44.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:25:53.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_12/index_4650_seed_22608_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:26:15.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 21:37:07.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_12/index_4927_seed_6093_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 21:38:30.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:30.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:36.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_12/index_4927_seed_6094_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:39:58.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:48.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 21:50:48.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_12/index_3751_seed_9068_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 21:52:09.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:52:09.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:21.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_12/index_3751_seed_9069_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:53:43.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 22:04:31.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_12/index_4886_seed_8168_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 22:05:51.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:51.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:04.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_12/index_4886_seed_8169_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:07:26.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.766 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 22:18:53.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_12/index_808_seed_12290_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 22:20:18.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:20:18.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:24.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_12/index_808_seed_12291_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:21:47.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 22:32:40.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_12/index_4622_seed_30853_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 22:34:00.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:34:00.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:11.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_12/index_4622_seed_30854_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:35:35.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 22:46:24.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_12/index_4523_seed_5668_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 22:47:44.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:44.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:48:56.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_12/index_4523_seed_5669_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:49:18.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:06.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 23:00:07.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_12/index_418_seed_17390_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 23:01:27.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:27.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:38.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_12/index_418_seed_17391_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:02:59.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 23:13:49.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:55.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_12/index_4234_seed_29705_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 23:15:10.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:15:10.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:20.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_12/index_4234_seed_29706_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:16:44.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 23:27:32.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_12/index_463_seed_32737_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 23:28:57.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:57.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:02.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_12/index_463_seed_32738_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:30:23.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 23:41:14.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_12/index_822_seed_7398_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 23:42:32.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:32.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:43:44.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_12/index_822_seed_7399_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 23:44:05.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:44:06.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-28 23:55:00.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_12/index_2826_seed_31017_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-28 23:56:18.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:18.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:28.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_12/index_2826_seed_31018_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:57:49.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-29 00:08:39.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_12/index_9960_seed_14884_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-29 00:10:00.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:10:00.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:08.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_12/index_9960_seed_14885_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:11:29.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-29 00:22:19.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_12/index_7222_seed_2586_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-29 00:23:39.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:39.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:24:52.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_12/index_7222_seed_2587_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:25:13.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.304 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-29 00:36:43.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_12/index_9330_seed_31146_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-29 00:38:02.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:38:02.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:15.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_12/index_9330_seed_31147_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:39:39.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-29 00:50:27.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_12/index_1412_seed_13469_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-29 00:51:47.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:47.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:52:56.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_12/index_1412_seed_13470_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:53:19.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 50, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 50, dp_rank 25: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=12, rank_in_group=1 +2026-06-29 01:04:09.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_12/index_3710_seed_29957_dp_25_rank_50_sample_2_slot_0.mp4 +2026-06-29 01:05:30.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 50 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:30.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 50 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:06:40.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 50: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_12/index_3710_seed_29958_dp_25_rank_50_sample_3_slot_1.mp4 +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 50: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 01:07:03.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 50: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 50: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 50, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_57.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_57.log new file mode 100644 index 0000000000000000000000000000000000000000..d1915e1aef92d54d7ef367633b04cc7b67733fbd --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_57.log @@ -0,0 +1,5147 @@ +2026-05-29 20:51:27.548 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:51:27.549 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 20:51:46.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:12.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 57, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 21:06:57.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:02.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 21:21:17.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:15.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 21:35:35.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:43.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 21:49:56.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:04.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 22:04:18.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 22:18:35.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:41.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 22:32:59.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:01.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 22:47:20.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:28.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 23:01:44.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:52.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:49.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:55.718 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 23:16:57.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:58.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 23:31:14.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 23:45:29.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:35.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:36.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-29 23:59:50.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:50.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:57.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 00:14:13.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:35.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 00:28:36.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:45.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 00:42:58.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:06.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 00:57:15.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:21.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:21.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 01:11:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 01:26:05.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.428 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:22.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:12.132 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 01:41:19.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:28.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:19.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 01:55:38.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:44.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:39.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 02:09:58.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:02.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 02:24:23.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:31.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 02:38:56.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:02.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:02.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:58.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:16.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 02:54:31.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:37.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:37.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 03:08:50.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:57.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:54.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 03:23:09.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:16.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:16.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 03:37:26.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 03:51:59.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:06.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:16.669 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 04:07:19.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 04:21:48.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:58.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 04:36:12.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 04:50:41.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:06.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 05:05:07.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:13.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:13.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 05:19:25.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 05:33:38.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:44.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:44.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:37.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 05:47:59.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:59.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 06:02:21.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.797 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 06:16:55.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 06:20:00.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:14.451 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 06:32:17.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:16.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 06:46:37.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:43.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:38.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 07:00:59.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:07.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:01.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 07:15:21.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:42.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 07:30:08.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 07:44:30.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:38.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 07:59:00.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:09.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 08:13:34.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:41.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 08:28:15.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:23.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:23.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:24.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 08:42:53.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:02.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:27.032 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 08:59:29.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:39.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:32.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 09:14:05.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:11.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:17.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 09:28:42.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:50.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:50.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 09:43:17.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:25.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 09:57:46.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:53.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 10:12:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:07.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 10:26:24.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:30.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:30.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:45.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 10:40:46.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:54.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 10:55:08.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:09.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:16.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:16.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 11:09:40.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:49.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.811 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:04.654 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 11:25:08.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:17.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:20.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 11:39:42.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:50.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.711 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 11:54:06.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 12:08:30.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:36.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:36.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:30.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 12:22:53.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:01.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:01.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 12:37:12.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:10.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 12:51:30.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 13:05:59.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:05.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:05.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:04.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 13:20:24.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:31.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:31.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:30.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 13:34:48.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:57.270 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 13:50:02.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:10.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:10.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 14:04:33.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 14:19:43.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:55.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 14:34:38.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.779 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 14:49:20.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:27.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 15:05:01.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:09.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:03.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 15:19:35.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 15:34:11.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:20.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:20.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 15:48:48.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:51.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 16:03:17.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:25.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.645 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 16:18:48.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:00.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:00.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 16:33:34.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 16:48:15.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.873 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 17:02:55.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 17:17:35.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:43.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:43.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 17:32:10.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 17:46:54.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:02.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:02.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 18:01:35.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 18:16:18.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:26.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 18:31:18.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:18.136 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 18:47:24.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:39.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 19:02:13.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:34.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 19:17:12.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:26.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 19:31:55.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 19:47:03.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:11.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:11.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 20:01:46.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 20:16:29.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 20:31:19.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 20:46:03.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:11.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:11.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 21:00:47.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:47.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:24.937 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 21:17:31.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:40.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:40.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:37.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 21:32:09.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:48:53.974 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:48:53.978 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 21:55:22.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:50.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:50.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:08.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 57, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 22:10:52.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:11:59.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 22:25:33.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:41.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:41.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 22:40:03.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:04.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:11.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:11.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:10.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 22:54:38.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:46.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:50.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 23:09:20.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:28.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 23:24:05.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:14.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 23:38:46.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:54.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.948 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-30 23:53:10.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:18.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:15.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 00:07:41.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:52.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:52.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:53.961 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 00:22:59.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 00:37:30.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:39.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 00:52:07.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:14.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:22.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 01:06:44.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:52.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:52.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 01:21:16.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 01:35:36.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:44.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 01:50:03.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:10.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:06.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 02:04:34.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:42.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:39.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 57, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-05-31 02:19:01.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:08.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:02.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:08.111 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:08.112 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 18:04:09.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:29.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:29.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 57, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 18:18:41.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 18:32:26.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 18:46:08.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:13.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:03.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 19:00:02.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 19:13:49.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 19:27:32.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:24.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 19:41:16.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:16.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:22.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.975 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 19:55:08.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:14.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.17s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 20:09:02.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:30.068 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 20:23:32.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 20:37:18.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:24.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 20:51:03.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 21:04:50.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:56.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 21:18:43.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:48.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.189 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 21:32:31.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 21:46:16.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 22:00:05.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:10.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 22:13:50.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:55.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:55.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 22:27:39.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:45.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:45.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:57.812 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 22:41:59.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 22:55:45.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:50.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 23:09:30.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:35.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 23:23:15.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:20.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:10.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 23:37:07.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:12.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:02.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-25 23:51:52.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 00:05:36.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 00:19:21.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:26.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 00:33:07.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 00:47:03.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:00.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:28.995 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 01:01:30.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 01:15:17.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:17.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:23.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:11.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 01:29:08.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:13.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:13.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 01:42:58.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.395 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 01:56:44.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 02:10:31.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 02:24:16.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.246 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 02:38:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:13.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 02:40:59.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 02:51:53.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:59.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 03:05:40.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:46.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.773 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 03:20:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 03:33:52.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:57.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:57.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 03:47:40.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:45.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 04:01:21.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 04:15:09.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 04:29:03.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 04:42:46.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 04:56:27.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:33.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 05:10:10.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:04.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 05:24:03.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.196 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 05:39:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:35.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 05:53:10.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:15.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:15.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 06:06:52.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 06:20:39.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:45.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 06:34:21.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:22.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:26.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:26.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 06:48:09.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:14.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 07:02:03.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:55.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 07:15:48.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 07:29:32.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:37.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:31.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 07:43:21.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:44.055 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 07:57:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:52.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 08:11:33.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 08:25:18.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 08:39:03.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:56.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 08:52:50.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:54.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:54.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 09:06:39.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 09:20:22.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:27.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:27.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 09:34:09.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:14.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 09:47:56.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:57.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:02.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:50.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 10:01:45.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:34.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.937 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 10:16:01.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:58.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 10:29:52.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:43.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 10:43:37.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:28.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 10:57:21.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 11:11:08.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:13.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:13.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 11:26:08.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:13.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.143 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 11:39:55.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 11:53:37.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:42.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:30.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 12:07:21.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:11.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 12:21:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:07.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:12.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:29.026 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 12:35:30.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:31.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:37.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:37.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 12:49:16.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:21.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:09.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 13:03:03.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:08.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:52.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 13:16:44.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 13:30:31.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 13:44:15.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 13:58:07.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 14:11:49.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:38.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 14:25:28.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:33.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:25.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 14:39:15.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:21.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:12.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.461 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 14:53:45.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 15:07:27.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 15:21:11.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:16.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 15:34:56.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 15:48:45.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:50.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 16:02:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:31.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:31.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:17.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 16:16:06.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:07.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:11.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:11.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:03.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.423 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 16:29:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:30:59.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:30:59.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 16:43:38.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.655 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 16:57:20.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:25.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:25.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.003 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 17:12:52.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:47.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 17:26:39.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:44.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 17:40:20.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 17:54:06.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 18:07:49.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 18:21:38.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:36.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 18:35:27.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:32.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:32.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 18:49:15.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:19.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:19.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:04.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.77s (1.88min) +2026-06-26 18:53:58.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 19:03:01.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:07.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 19:16:46.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:51.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.825 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 19:31:08.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:06.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 19:45:03.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 19:58:45.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:45.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:51.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:37.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 20:12:28.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:21.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 20:26:14.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 20:40:07.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:11.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:11.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 20:53:52.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:58.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:53.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 21:07:44.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 21:21:31.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 21:35:16.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:16.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:21.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:21.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:10.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.672 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 21:49:43.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 22:03:26.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:27.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:32.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:17.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 22:17:07.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 22:30:49.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 22:44:32.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 22:59:16.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 23:13:02.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:08.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:08.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:52.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 23:26:44.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 23:40:27.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:25.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-26 23:54:16.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:21.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.235 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 00:08:40.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 00:22:20.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 00:36:06.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 00:49:49.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 01:03:35.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:40.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 01:17:19.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:24.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:24.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 01:31:06.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:12.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:58.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 01:44:47.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:52.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:52.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 01:58:27.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 02:12:14.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.762 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 02:26:39.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:46.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:33.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 02:40:23.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:28.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 02:54:14.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:09.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 03:08:01.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:07.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:49.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 03:21:41.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:46.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:46.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 03:35:26.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 03:49:14.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 04:03:03.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:08.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 04:16:48.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:53.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 04:30:30.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:35.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:35.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.467 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 04:45:58.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:05.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 04:59:46.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:52.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:39.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 05:13:27.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:32.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:32.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 05:27:07.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:08.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:12.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:12.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 05:40:50.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 05:54:35.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:40.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 06:08:19.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:13.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 06:22:07.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:07.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:12.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:12.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 06:35:50.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:55.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 06:49:30.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:30.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:48.007 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 07:03:49.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:55.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:56.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 07:17:29.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:34.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:22.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 07:31:16.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:11.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 07:45:07.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 07:58:52.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:57.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 08:12:38.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:42.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:42.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 08:26:20.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:10.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 08:40:03.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:07.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:08.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 08:53:47.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:53.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 09:07:41.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:46.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.246 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 09:22:02.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 09:35:43.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:47.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:47.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 09:49:26.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:31.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:17.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 10:03:08.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:12.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:12.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 10:16:53.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:57.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:57.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 10:31:49.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:54.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 10:45:31.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 10:59:16.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:21.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:21.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:12.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 11:13:07.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:12.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.423 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 11:26:47.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:41.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.617 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 11:41:06.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:11.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:11.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:58.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 11:54:48.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:52.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:52.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 12:08:34.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:39.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.701 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 12:22:19.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 12:36:01.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 12:49:44.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 13:03:21.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 13:17:08.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:13.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 13:30:47.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:46.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 13:44:38.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:43.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.218 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 13:58:57.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:02.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.484 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 14:12:43.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 14:26:20.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:25.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:11.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 14:40:05.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:06.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:10.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:10.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:58.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 14:53:47.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 15:07:27.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:31.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:31.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 15:21:06.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:11.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:58.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 15:34:47.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:47.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:53.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 15:48:32.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:36.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:36.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 16:02:14.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:05.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.810 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 16:17:43.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:44.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:49.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:49.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 16:31:26.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 16:45:07.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:59.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 16:58:47.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:52.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:42.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 17:12:31.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 17:26:15.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:20.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:09.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 17:40:00.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:00.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:05.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 17:53:45.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:38.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 18:07:25.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:24.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 18:21:14.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:19.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:19.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:07.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.569 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.689 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 18:35:38.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:43.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.262 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 18:49:21.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 19:03:06.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:11.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 19:16:46.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:51.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 19:30:30.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:25.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 19:44:14.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:19.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:10.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 19:58:02.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:53.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 20:11:43.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:48.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:48.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:38.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 20:25:26.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:19.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 20:39:06.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:07.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:12.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.555 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 20:53:29.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:35.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:35.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.307 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:21.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 21:07:10.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:15.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:06.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 21:21:01.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:51.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 21:34:42.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 21:48:23.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 22:03:21.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:16.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 22:17:06.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:07.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:57.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 22:30:47.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:52.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 22:44:30.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 22:58:07.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:12.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.522 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 23:12:29.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:35.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:35.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 23:26:15.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:19.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:20.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.220 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 23:40:02.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:50.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-27 23:53:51.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:56.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.975 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 00:07:32.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.021 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 00:21:21.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:21.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:26.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:26.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 00:35:07.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 00:49:02.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:07.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:56.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 01:02:48.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 01:16:33.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.817 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 01:30:58.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 01:44:44.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 01:58:27.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 02:12:08.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:13.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:04.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 02:25:53.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 02:39:38.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:31.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 02:53:20.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:25.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:25.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 03:07:08.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:03.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 03:20:53.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:49.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 03:34:40.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:45.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.757 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 03:50:13.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 04:04:03.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 04:17:48.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:53.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:53.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 04:31:40.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 04:45:20.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:11.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 04:59:03.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 05:12:44.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:49.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:49.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:35.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.188 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 05:26:26.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:22.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 05:40:15.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 05:54:07.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:07.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:12.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:25.051 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 06:08:27.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:20.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 06:22:11.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:17.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.326 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 06:36:02.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:07.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:44.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 06:49:45.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 07:03:27.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:28.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 07:17:12.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:17.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:04.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:53.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 07:30:53.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:48.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 07:44:40.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:33.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 07:58:21.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:26.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:26.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 08:12:08.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:13.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.458 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 08:26:31.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 08:40:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:23.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:28.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:13.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 08:54:08.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:13.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 08:56:59.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 09:07:50.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:55.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:55.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 09:21:32.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:37.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 09:36:35.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 09:50:22.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:24.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 10:04:16.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 10:18:09.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:14.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:02.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 10:31:54.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:53.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:19.104 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 10:46:20.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 11:00:08.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 11:13:51.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 11:27:32.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 11:41:15.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 11:55:03.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 12:08:48.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:53.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:47.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 12:22:40.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 12:36:24.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 12:50:08.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:13.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:13.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.487 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 13:04:33.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:34.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 13:18:22.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:27.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:27.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:14.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 13:32:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 13:45:49.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:49.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:54.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:54.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 13:59:30.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:30.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:35.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:35.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:21.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 14:13:08.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 14:27:02.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 14:40:44.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:48.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:48.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 14:54:25.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 15:08:07.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:12.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:02.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.811 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.353 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 15:23:36.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:31.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 15:37:21.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 15:51:08.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 16:04:54.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:49.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 16:18:40.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:41.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:45.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:45.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.211 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 16:32:23.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:28.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 16:46:07.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 16:59:50.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 17:13:34.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 17:27:21.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:26.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.767 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 17:41:44.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:49.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:49.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 17:55:32.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:32.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 18:09:16.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:21.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 18:23:03.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 18:36:44.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:44.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:49.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:49.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:38.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 18:50:26.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:22.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 19:04:13.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:14.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:19.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:07.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 19:18:02.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:06.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:06.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 19:31:44.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 19:45:25.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:29.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:29.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.550 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.410 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 19:59:44.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 20:13:24.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:29.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:29.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:15.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 20:27:07.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:12.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 20:40:49.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:54.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 20:54:36.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 21:09:41.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:46.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:46.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 21:23:23.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:29.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 21:37:07.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:48.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 21:50:48.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:53.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:53.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 22:04:32.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:36.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:36.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.672 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 22:18:53.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 22:32:40.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 22:46:24.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:29.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 23:00:07.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:59.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 23:13:49.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:54.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 23:27:32.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:23.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 23:41:14.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.373 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-28 23:55:00.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-29 00:08:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:28.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-29 00:22:19.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.994 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.354 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-29 00:36:43.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:48.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-29 00:50:27.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.326 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.464 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 57, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 57, dp_rank 28: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=0 +2026-06-29 01:04:09.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 57 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:14.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 57 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 57: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:03.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 57: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 57: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 57, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_59.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_59.log new file mode 100644 index 0000000000000000000000000000000000000000..cb268a6a2e263c0f2fb3c2aaa5769d61bfe99cb2 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_59.log @@ -0,0 +1,5147 @@ +2026-05-29 20:51:25.739 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:51:25.740 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 20:51:46.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:12.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:12.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 59, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 21:06:57.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:02.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 21:21:17.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:16.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.878 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 21:35:35.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:43.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 21:49:56.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:04.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.293 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 22:04:18.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 22:18:35.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:42.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:42.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 22:32:58.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:01.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 22:47:20.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:28.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:28.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 23:01:44.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:52.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:49.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:54.515 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 23:16:57.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:58.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:13.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 23:31:14.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.281 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 23:45:29.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:50.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-29 23:59:50.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:50.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:57.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 00:14:13.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 00:28:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:45.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 00:42:58.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:06.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 00:57:15.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:22.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 01:11:38.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 01:26:05.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.428 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:17.091 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 01:41:19.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:28.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:28.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:19.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 01:55:38.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:45.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:45.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:40.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.142 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 02:09:58.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:02.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 02:24:23.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:31.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 02:38:56.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:03.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:03.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:58.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:16.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 02:54:31.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:38.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 03:08:50.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:57.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:53.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 03:23:09.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:17.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:17.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.523 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 03:37:26.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 03:51:59.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:06.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:17.709 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 04:07:19.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 04:21:48.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:58.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 04:36:12.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 04:50:41.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 05:05:07.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:13.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:13.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 05:19:25.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 05:33:38.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:45.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:45.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:38.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 05:47:59.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:58.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 06:02:21.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 06:16:55.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 06:19:59.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:14.368 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 06:32:17.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:17.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 06:46:37.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:43.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:38.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 07:00:59.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:07.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:01.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 07:15:22.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:22.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:42.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 07:30:08.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 07:44:30.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:38.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:38.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:59.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 07:59:00.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:09.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 08:13:34.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:41.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 08:28:15.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:23.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:23.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:24.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 08:42:53.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:02.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:27.288 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 08:59:29.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:39.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:33.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.731 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 09:14:05.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:11.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:17.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.663 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 09:28:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:51.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 09:43:17.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:25.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 09:57:46.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:54.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:54.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 10:12:07.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:07.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 10:26:24.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:30.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:30.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.331 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:45.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 10:40:46.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:54.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 10:55:08.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:17.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 11:09:40.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:49.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:04.123 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 11:25:07.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:17.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:21.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 11:39:42.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:50.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 11:54:06.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 12:08:30.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:37.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:30.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 12:22:53.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:01.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:01.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 12:37:12.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:11.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 12:51:30.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.999 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 13:05:59.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:05.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:05.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:04.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 13:20:24.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:31.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:31.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:30.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 13:34:48.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:57.988 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 13:50:02.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:11.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 14:04:33.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 14:19:43.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:55.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:55.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 14:34:38.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 14:49:20.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:27.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 15:05:01.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:09.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:04.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 15:19:35.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.731 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 15:34:11.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:20.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:20.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 15:48:48.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:51.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 16:03:17.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:25.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.387 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 16:18:48.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:00.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:00.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 16:33:34.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 16:48:15.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 17:02:55.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 17:17:35.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:44.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 17:32:10.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 17:46:54.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:02.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:02.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 18:01:35.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 18:16:18.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:26.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 18:31:18.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:19.375 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 18:47:24.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:39.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 19:02:13.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:34.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 19:17:12.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:13.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:26.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 19:31:55.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 19:47:03.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:12.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 20:01:46.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 20:16:29.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 20:31:20.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 20:46:03.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:11.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:12.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 21:00:47.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:24.979 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 21:17:31.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:40.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:40.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:37.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 21:32:08.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:08.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:48:56.953 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:48:56.954 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 21:55:22.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:50.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:50.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:08.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 59, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 22:10:52.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:11:59.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 22:25:33.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:42.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:42.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 22:40:03.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:11.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:11.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:10.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 22:54:38.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:38.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:46.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:49.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 23:09:20.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:28.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:28.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 23:24:05.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:14.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 23:38:46.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:54.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-30 23:53:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:18.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:15.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 00:07:41.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:52.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:52.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:54.437 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 00:22:59.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 00:37:30.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:39.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 00:52:07.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:14.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:21.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 01:06:44.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:52.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:52.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 01:21:16.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 01:35:36.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:44.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 01:50:03.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:10.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:06.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 02:04:34.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:42.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:39.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 59, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-05-31 02:19:01.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:08.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:02.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:07.895 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:07.896 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 18:04:09.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:29.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:29.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 59, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 18:18:41.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:35.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 18:32:26.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 18:46:08.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:12.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:12.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:04.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 19:00:02.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:02.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 19:13:49.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 19:27:32.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:24.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 19:41:17.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:21.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:21.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 19:55:08.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:14.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.17s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 20:09:02.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.941 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 20:23:32.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 20:37:18.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:19.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:23.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:23.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 20:51:03.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 21:04:50.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:56.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 21:18:43.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:48.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.189 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 21:32:31.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 21:46:16.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 22:00:05.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:10.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:10.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 22:13:50.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:55.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:55.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 22:27:39.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:45.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:45.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:57.768 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 22:41:59.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 22:55:45.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:50.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 23:09:30.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:35.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 23:23:15.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:20.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:11.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 23:37:07.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:11.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:11.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:02.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-25 23:51:52.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 00:05:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.385 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 00:19:21.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:25.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:25.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 00:33:07.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 00:47:03.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:00.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:28.768 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 01:01:30.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 01:15:17.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:23.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:12.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 01:29:08.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:14.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:14.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 01:42:58.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 01:56:44.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 02:10:31.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 02:24:16.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 02:38:08.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:12.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:13.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 02:41:00.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 02:51:53.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:54.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:59.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 03:05:40.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:41.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:45.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:46.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.665 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 03:20:06.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 03:33:52.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:57.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:57.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 03:47:40.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:45.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 04:01:21.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.617 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 04:15:09.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 04:29:02.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 04:42:46.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.439 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 04:56:27.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:33.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.652 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 05:10:10.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:04.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 05:24:03.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:25.970 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 05:39:29.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:34.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:34.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 05:53:10.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:15.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:15.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 06:06:52.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 06:20:40.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:44.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 06:34:21.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:26.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:26.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 06:48:09.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:14.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 07:02:03.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:55.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 07:15:48.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 07:29:32.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:36.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:36.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:30.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 07:43:21.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.859 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 07:57:45.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:45.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:52.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 08:11:33.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 08:25:18.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:08.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 08:39:03.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:57.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 08:52:50.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:54.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:54.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 09:06:39.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 09:20:22.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:27.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:27.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 09:34:09.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:14.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 09:47:56.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:02.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:51.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 10:01:45.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.543 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 10:16:01.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:58.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 10:29:52.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:44.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 10:43:37.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:29.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 10:57:21.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 11:11:08.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:09.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:13.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:13.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 11:26:08.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:13.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.245 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 11:39:55.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 11:53:37.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:41.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:41.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:31.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 12:07:21.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:12.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:05.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 12:21:06.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:12.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.786 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 12:35:30.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:37.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:37.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 12:49:16.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:21.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:09.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.064 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 13:03:03.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:08.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:53.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 13:16:44.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:49.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 13:30:31.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 13:44:15.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.616 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 13:58:07.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 14:11:49.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:39.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 14:25:28.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:33.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:25.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 14:39:15.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:20.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:20.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:13.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.586 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 14:53:45.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 15:07:27.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 15:21:10.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:16.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.252 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 15:34:56.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 15:48:45.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:50.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 16:02:26.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:31.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:31.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:17.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:05.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 16:16:06.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:11.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:11.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:04.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 16:29:54.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:30:59.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:30:59.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 16:43:38.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.656 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 16:57:20.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:25.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:25.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.285 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 17:12:53.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:48.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 17:26:39.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:44.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 17:40:20.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 17:54:06.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 18:07:49.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 18:21:38.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:37.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.994 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 18:35:27.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:32.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:33.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 18:49:15.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:20.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:05.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 19:03:02.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:07.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 19:16:46.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:51.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.945 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 19:31:08.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:06.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 19:45:03.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 19:58:46.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:51.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:37.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 20:12:28.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:21.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 20:26:14.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 20:40:06.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:11.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:11.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 20:53:52.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:58.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:54.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 21:07:44.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 21:21:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 21:35:16.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:21.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:21.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:11.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.279 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 21:49:43.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 22:03:26.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:31.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:31.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:17.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 22:17:07.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 22:30:49.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 22:44:32.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 22:59:16.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 23:13:02.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:07.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:07.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:53.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 23:26:44.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 23:40:27.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:26.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-26 23:54:16.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:21.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.172 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 00:08:40.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 00:22:20.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 00:36:06.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:59.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 00:49:49.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 01:03:35.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:39.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:39.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 01:17:19.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:20.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:24.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:24.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 01:31:06.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:06.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:11.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:11.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:59.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 01:44:47.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:52.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:52.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 01:58:27.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 02:12:14.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.569 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 02:26:39.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:46.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:33.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 02:40:23.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:28.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 02:54:14.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:09.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 03:08:01.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:02.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:07.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:50.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 03:21:41.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:46.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:46.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 03:35:26.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 03:49:14.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 04:03:03.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:08.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 04:16:48.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:53.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 04:30:31.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:36.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.527 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 04:45:58.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:04.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:04.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 04:59:46.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:52.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:39.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 05:13:27.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:27.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:32.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:32.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 05:27:07.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:12.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:12.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 05:40:50.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 05:54:35.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:40.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 06:08:19.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:13.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 06:22:07.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:12.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:12.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 06:35:50.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:55.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 06:49:31.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.789 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 07:03:49.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:55.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:55.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 07:17:29.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:34.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:22.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 07:31:16.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:11.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 07:45:07.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 07:58:52.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:57.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 08:12:38.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:42.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:42.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 08:26:20.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:10.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 08:40:03.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:08.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 08:53:47.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:53.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 09:07:41.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:46.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.355 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 09:22:02.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 09:35:43.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:47.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:47.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 09:49:26.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:31.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:17.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 10:03:08.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:12.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:12.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.863 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 10:16:53.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:57.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:57.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 10:31:49.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:49.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:54.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 10:45:31.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 10:59:16.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:21.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:21.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:12.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 11:13:07.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:11.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:11.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 11:26:47.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:41.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.749 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 11:41:06.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:11.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:11.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:58.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 11:54:48.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:52.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:52.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 12:08:34.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:38.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:38.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 12:22:19.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 12:36:01.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 12:49:44.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 13:03:21.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 13:17:08.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:12.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:13.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 13:30:47.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:47.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 13:44:38.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:43.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.594 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 13:58:57.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:02.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 14:12:43.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 14:26:20.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:25.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:12.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:04.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 14:40:05.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:06.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:10.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:10.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:57.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 14:53:47.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 15:07:26.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:31.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 15:21:06.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:11.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:57.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 15:34:48.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:52.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:52.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 15:48:32.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:36.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:36.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 16:02:14.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:04.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:42.172 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 16:17:43.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:44.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:49.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:49.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 16:31:26.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 16:45:07.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:58.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.780 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 16:58:47.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:52.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:41.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 17:12:31.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 17:26:15.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:20.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:08.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 17:40:00.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:05.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.806 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 17:53:45.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:37.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 18:07:25.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:23.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 18:21:14.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:19.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:19.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:07.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.748 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 18:35:38.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:43.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.246 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 18:49:21.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 19:03:06.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:11.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 19:16:46.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:50.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:51.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 19:30:30.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:24.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 19:44:14.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:18.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:18.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:09.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 19:58:02.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:53.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 20:11:44.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:48.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:48.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:38.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 20:25:26.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:18.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 20:39:06.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:12.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.548 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 20:53:29.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:34.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.307 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:21.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 21:07:10.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:15.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:06.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 21:21:01.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:51.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 21:34:42.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 21:48:24.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 22:03:21.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:16.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 22:17:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:07.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:56.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 22:30:47.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:52.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 22:44:30.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 22:58:07.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:11.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:11.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.894 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 23:12:29.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:35.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:35.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 23:26:15.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:19.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:19.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 23:40:02.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-27 23:53:51.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:56.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 00:07:32.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 00:21:21.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:26.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:26.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 00:35:07.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 00:49:02.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:07.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:56.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.010 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 01:02:48.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 01:16:33.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.957 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 01:30:58.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 01:44:44.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 01:58:27.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 02:12:08.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:13.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:03.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.550 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 02:25:53.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 02:39:38.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:30.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 02:53:20.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:25.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:25.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 03:07:08.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:02.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 03:20:53.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:48.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 03:34:40.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:45.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:12.136 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 03:50:13.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 04:04:03.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 04:17:48.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:53.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:53.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 04:31:40.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 04:45:20.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:10.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 04:59:03.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 05:12:44.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:49.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:49.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:34.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 05:26:26.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:21.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 05:40:15.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 05:54:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:07.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:11.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:11.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:25.234 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 06:08:27.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:21.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 06:22:11.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:12.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:17.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:06.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 06:36:02.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:07.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:44.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 06:49:45.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 07:03:27.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 07:17:12.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:16.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:16.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:04.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 07:30:53.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:47.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 07:44:40.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:32.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 07:58:21.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:26.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:26.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 08:12:08.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:13.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:13.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.800 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 08:26:31.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 08:40:22.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:28.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:13.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 08:54:08.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:12.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:12.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 08:56:59.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 09:07:50.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:51.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:55.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:55.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 09:21:32.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:37.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.696 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 09:36:35.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 09:50:22.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:24.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 10:04:16.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 10:18:09.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:14.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:01.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 10:31:54.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:53.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:19.055 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 10:46:20.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 11:00:08.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 11:13:51.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 11:27:32.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 11:41:15.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 11:55:03.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 12:08:48.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:53.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:47.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 12:22:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 12:36:24.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 12:50:08.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:09.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:13.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:13.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.832 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 13:04:33.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:33.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 13:18:22.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:27.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:27.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:14.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 13:32:07.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.092 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 13:45:50.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:54.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:54.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 13:59:31.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:35.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:35.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:20.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 14:13:08.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 14:27:02.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 14:40:44.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:48.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:48.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 14:54:25.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 15:08:07.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:11.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:11.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:01.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.061 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 15:23:36.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:31.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 15:37:21.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 15:51:08.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 16:04:54.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:50.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 16:18:41.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:41.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:45.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:45.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 16:32:23.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 16:46:07.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.395 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 16:59:50.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 17:13:34.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 17:27:21.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:25.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:25.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:43.094 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 17:41:44.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:49.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:49.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 17:55:31.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:32.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 18:09:16.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:20.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:20.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 18:23:03.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 18:36:45.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:49.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:49.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:39.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 18:50:26.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:22.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 19:04:13.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:19.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:07.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 19:18:01.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:06.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:06.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 19:31:44.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 19:45:24.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:24.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:29.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:29.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.308 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 19:59:44.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 20:13:24.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:24.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:29.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:29.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:15.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 20:27:07.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:11.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:11.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 20:40:49.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:54.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 20:54:36.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 21:09:41.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:46.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:46.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 21:23:24.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:29.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 21:37:07.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 21:50:48.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:49.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:53.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 22:04:31.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:36.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:36.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.735 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 22:18:53.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 22:32:40.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:34.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 22:46:24.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:28.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:28.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 23:00:07.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:58.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 23:13:49.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:54.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 23:27:32.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:22.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 23:41:14.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-28 23:55:00.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-29 00:08:39.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:28.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-29 00:22:19.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.994 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.213 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.151 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-29 00:36:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:47.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:47.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-29 00:50:27.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 59, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 59, dp_rank 29: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=14, rank_in_group=1 +2026-06-29 01:04:09.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 59 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:14.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 59 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 59: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:03.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 59: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 59: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 59, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_62.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_62.log new file mode 100644 index 0000000000000000000000000000000000000000..560c9e9f60117592daee46b4399097f4675c14e6 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_62.log @@ -0,0 +1,6079 @@ +2026-05-29 20:51:42.404 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-29 20:51:42.404 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-29 20:51:43.204 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:51:43.205 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 20:51:46.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:13.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_15/index_717_seed_14730_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 20:53:32.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:32.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:39.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_15/index_717_seed_14731_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 20:55:17.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 62, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.569 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 21:06:57.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:05.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_15/index_9841_seed_6830_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 21:08:22.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:22.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:09:39.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_15/index_9841_seed_6831_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 21:10:00.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:10:02.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 21:21:17.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:25.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_15/index_6303_seed_21750_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 21:22:44.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:44.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:23:53.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_15/index_6303_seed_21751_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:24:16.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 21:35:35.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:44.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_15/index_6605_seed_20968_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 21:37:05.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:37:05.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:14.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_15/index_6605_seed_20969_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:38:36.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 21:49:56.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:05.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_15/index_2454_seed_12211_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 21:51:27.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:27.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:52:36.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_15/index_2454_seed_12212_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 21:53:01.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 22:04:18.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_15/index_8510_seed_13058_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 22:05:45.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:45.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:06:53.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_15/index_8510_seed_13059_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:07:15.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 22:18:35.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:43.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_15/index_4287_seed_30260_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 22:20:00.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:20:00.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:15.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_15/index_4287_seed_30261_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:21:38.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 22:32:59.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:06.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_15/index_8181_seed_26292_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 22:34:29.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:29.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:36.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_15/index_8181_seed_26293_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:36:01.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 22:47:20.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:29.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_15/index_9102_seed_26354_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 22:48:48.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:48.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:49:57.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_15/index_9102_seed_26355_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 22:50:20.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 23:01:44.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:52.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_15/index_4081_seed_18322_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 23:03:17.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:03:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:24.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_15/index_4081_seed_18323_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:04:49.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:54.844 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 23:16:57.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:06.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_15/index_7559_seed_13217_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 23:18:25.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:25.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:35.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_15/index_7559_seed_13218_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 23:19:57.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:19:58.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:13.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 23:31:14.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:21.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_15/index_2609_seed_17043_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 23:32:39.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:39.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:33:53.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_15/index_2609_seed_17044_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:34:15.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 23:45:29.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_15/index_1478_seed_28855_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-29 23:46:58.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:58.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:09.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_15/index_1478_seed_28856_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-29 23:48:31.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-29 23:59:50.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:51.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:58.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_15/index_3403_seed_21541_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 00:01:19.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:01:19.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:27.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_15/index_3403_seed_21542_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:02:55.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 00:14:13.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:20.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_15/index_626_seed_13193_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 00:15:42.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:42.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:16:51.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_15/index_626_seed_13194_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:17:18.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 00:28:36.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_15/index_9178_seed_3521_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 00:30:02.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:30:02.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:17.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_15/index_9178_seed_3522_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 00:31:38.351 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:31:39.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 00:42:58.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:07.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_15/index_7279_seed_10954_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 00:44:32.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:32.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:45:38.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_15/index_7279_seed_10955_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 00:46:02.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 00:57:15.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:22.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_15/index_2137_seed_4454_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 00:58:45.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:45.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:59:51.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_15/index_2137_seed_4455_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:00:17.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 01:11:38.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:46.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_15/index_1268_seed_19666_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 01:13:08.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:13:08.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:19.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_15/index_1268_seed_19667_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:14:44.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 01:26:05.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:15.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_15/index_2828_seed_2238_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 01:27:35.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:35.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:28:42.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_15/index_2828_seed_2239_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 01:29:03.428 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:29:05.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.014 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:22.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:12.062 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 01:41:19.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:29.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_15/index_7943_seed_2338_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 01:42:49.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:49.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:43:56.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_15/index_7943_seed_2339_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:44:19.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 01:55:38.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:45.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_15/index_324_seed_23323_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 01:57:03.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:57:03.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:15.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_15/index_324_seed_23324_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 01:58:40.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 02:09:58.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:05.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_15/index_3082_seed_14010_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 02:11:27.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:27.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:12:40.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_15/index_3082_seed_14011_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:13:03.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 02:24:23.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:32.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_15/index_7405_seed_7317_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 02:25:59.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:59.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:06.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_15/index_7405_seed_7318_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:27:33.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.541 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 02:38:56.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:04.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_15/index_2779_seed_26820_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 02:40:23.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:23.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_15/index_2779_seed_26821_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:41:58.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:16.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 02:54:31.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:38.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_15/index_5540_seed_8983_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 02:55:57.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:57.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:09.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_15/index_5540_seed_8984_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 02:57:31.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.301 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 03:08:50.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_15/index_1868_seed_8463_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 03:10:20.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:10:20.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:27.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_15/index_1868_seed_8464_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:11:54.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 03:23:09.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:18.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_15/index_921_seed_4158_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 03:24:34.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:34.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:25:47.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_15/index_921_seed_4159_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:26:08.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.524 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 03:37:26.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:27.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:34.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_15/index_9047_seed_7619_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 03:39:01.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:39:01.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:08.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_15/index_9047_seed_7620_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:40:39.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.984 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 03:51:59.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_15/index_3812_seed_19120_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 03:53:32.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:32.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:54:39.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_15/index_3812_seed_19121_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 03:55:04.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.947 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:17.252 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 04:07:19.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:29.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_15/index_144_seed_13716_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 04:08:53.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:53.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:01.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_15/index_144_seed_13717_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:10:28.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 04:21:48.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:23:00.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_15/index_6049_seed_13910_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 04:23:16.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:23:16.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:29.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_15/index_6049_seed_13911_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 04:24:53.586 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:24:54.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 04:36:12.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:20.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_15/index_9215_seed_1221_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 04:37:41.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:41.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:38:50.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_15/index_9215_seed_1222_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:39:16.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 04:50:41.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:48.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_15/index_3370_seed_27564_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 04:52:11.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:52:11.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:19.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_15/index_3370_seed_27565_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 04:53:44.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 05:05:07.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:14.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_15/index_9829_seed_23998_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 05:06:37.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:37.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:07:44.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_15/index_9829_seed_23999_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:08:13.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 05:19:25.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:32.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_15/index_3900_seed_28113_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 05:20:51.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:51.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:21:59.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_15/index_3900_seed_28114_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:22:21.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 05:33:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:46.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_15/index_9770_seed_8957_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 05:35:10.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:35:10.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:16.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_15/index_9770_seed_8958_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:36:37.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 05:47:59.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:07.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_15/index_2632_seed_26757_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 05:49:25.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:25.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:36.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_15/index_2632_seed_26758_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 05:50:57.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 05:50:59.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 06:02:21.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:30.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_15/index_2354_seed_7979_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 06:03:50.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:50.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:05.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_15/index_2354_seed_7980_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:05:33.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.826 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 06:16:55.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:04.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_15/index_1020_seed_28649_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 06:18:25.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:25.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:32.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_15/index_1020_seed_28650_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:20:00.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:14.092 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 06:32:17.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:26.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_15/index_5189_seed_24901_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 06:33:47.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:47.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:34:55.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_15/index_5189_seed_24902_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:35:17.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 06:46:37.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:44.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_15/index_2593_seed_24179_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 06:47:59.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:59.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:15.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_15/index_2593_seed_24180_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 06:49:38.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 07:00:59.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:08.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_15/index_655_seed_4248_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 07:02:24.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:24.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:38.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_15/index_655_seed_4249_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:04:01.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 07:15:21.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:29.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_15/index_124_seed_15797_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 07:16:49.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:49.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:06.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_15/index_124_seed_15798_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:18:42.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.029 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 07:30:08.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:17.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_15/index_1789_seed_31677_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 07:31:41.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:41.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:32:49.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_15/index_1789_seed_31678_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:33:11.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 07:44:30.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:39.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_15/index_5125_seed_1906_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 07:45:58.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:58.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:05.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_15/index_5125_seed_1907_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 07:47:31.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:59.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 07:59:00.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:08.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_15/index_852_seed_16014_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 08:00:31.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:31.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:01:42.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_15/index_852_seed_16015_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:02:08.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 08:13:34.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:41.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_15/index_5634_seed_4597_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 08:14:58.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:58.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:17.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_15/index_5634_seed_4598_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:16:44.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 08:28:15.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:24.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_15/index_3269_seed_28363_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 08:29:48.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:48.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:30:56.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_15/index_3269_seed_28364_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 08:31:22.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:31:24.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:52.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 08:42:53.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:03.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_15/index_1674_seed_30701_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 08:44:24.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:24.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:45:36.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_15/index_1674_seed_30702_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 08:46:01.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:26.953 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 08:59:29.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:41.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_15/index_4495_seed_30381_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 09:00:58.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:58.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:11.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_15/index_4495_seed_30382_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 09:02:32.038 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:02:33.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 09:14:05.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:13.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_15/index_3575_seed_28158_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 09:15:28.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:28.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:16:53.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_15/index_3575_seed_28159_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:17:17.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 09:28:42.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:52.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_15/index_7043_seed_8519_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 09:30:11.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:30:11.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:27.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_15/index_7043_seed_8520_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:31:54.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 09:43:17.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_15/index_9111_seed_21779_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 09:44:55.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:55.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:02.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_15/index_9111_seed_21780_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 09:46:24.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 09:46:25.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 09:57:46.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:54.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_15/index_1012_seed_7920_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 09:59:12.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:59:12.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:24.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_15/index_1012_seed_7921_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:00:52.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 10:12:07.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:16.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_15/index_3487_seed_14021_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 10:13:34.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:34.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:14:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_15/index_3487_seed_14022_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:15:06.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 10:26:24.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_15/index_4599_seed_25672_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 10:27:50.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:50.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:03.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_15/index_4599_seed_25673_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:29:29.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:45.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 10:40:46.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_15/index_6060_seed_15367_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 10:42:12.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:42:12.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:28.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_15/index_6060_seed_15368_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:43:55.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.639 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 10:55:08.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:17.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_15/index_1449_seed_23327_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 10:56:35.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:35.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:57:53.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_15/index_1449_seed_23328_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 10:58:23.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:39.211 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 11:09:40.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:49.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_15/index_4396_seed_27276_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 11:11:10.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:11:10.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:20.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_15/index_4396_seed_27277_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:12:43.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.894 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:04.644 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 11:25:08.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:18.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_15/index_3365_seed_5979_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 11:26:37.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:37.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:27:53.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_15/index_3365_seed_5980_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:28:21.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 11:39:42.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:51.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_15/index_1672_seed_23932_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 11:41:09.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:41:09.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:20.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_15/index_1672_seed_23933_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:42:42.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 11:54:06.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:14.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_15/index_6641_seed_21718_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 11:55:35.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:35.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:56:42.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_15/index_6641_seed_21719_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 11:57:06.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.841 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 12:08:30.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_15/index_3330_seed_7892_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 12:09:59.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:59.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:07.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_15/index_3330_seed_7893_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:11:30.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 12:22:53.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:02.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_15/index_3776_seed_26267_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 12:24:19.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:19.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:30.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_15/index_3776_seed_26268_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:25:55.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 12:37:12.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:20.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_15/index_6735_seed_18147_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 12:38:38.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:38.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:39:47.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_15/index_6735_seed_18148_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:40:11.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 12:51:30.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_15/index_1842_seed_9718_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 12:52:55.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:55.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:11.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_15/index_1842_seed_9719_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 12:54:32.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 13:05:59.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_15/index_3450_seed_7520_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 13:07:20.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:20.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:08:38.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_15/index_3450_seed_7521_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:09:04.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 13:20:24.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:33.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_15/index_779_seed_16678_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 13:21:54.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:54.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:00.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_15/index_779_seed_16679_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:23:30.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 13:34:48.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:56.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_15/index_5385_seed_11080_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 13:36:12.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:36:12.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:25.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_15/index_5385_seed_11081_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 13:37:47.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:37:48.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.960 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 13:50:02.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_15/index_3144_seed_24318_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 13:51:31.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:31.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:52:48.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_15/index_3144_seed_24319_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 13:53:11.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 14:04:33.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:44.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_15/index_1530_seed_11221_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 14:06:00.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:06:00.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:19.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_15/index_1530_seed_11222_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:07:59.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 14:19:43.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:56.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_15/index_151_seed_22664_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 14:21:10.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:21:10.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:22:39.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_15/index_151_seed_22665_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 14:23:12.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:23:13.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 14:34:38.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:46.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_15/index_9888_seed_14162_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 14:36:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:36:02.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:17.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_15/index_9888_seed_14163_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:37:54.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 14:49:20.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:28.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_15/index_713_seed_2943_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 14:50:40.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:40.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:00.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_15/index_713_seed_2944_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 14:52:26.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 15:05:01.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_15/index_8335_seed_7456_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 15:06:24.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:24.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:07:40.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_15/index_8335_seed_7457_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:08:04.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 15:19:35.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:43.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_15/index_8616_seed_20909_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 15:20:58.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:58.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:14.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_15/index_8616_seed_20910_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 15:22:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:22:38.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 15:34:11.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:21.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_15/index_8954_seed_12636_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 15:35:39.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:39.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:36:53.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_15/index_8954_seed_12637_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:37:19.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 15:48:48.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:57.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_15/index_4060_seed_5020_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 15:50:13.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:50:13.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:25.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_15/index_4060_seed_5021_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 15:51:49.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 15:51:51.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 16:03:17.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:26.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_15/index_2990_seed_13634_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 16:04:40.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:40.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:05:57.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_15/index_2990_seed_13635_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:06:20.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.701 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 16:18:48.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_15/index_3123_seed_443_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 16:20:18.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:18.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:21:33.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_15/index_3123_seed_444_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:22:05.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 16:33:33.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:43.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_15/index_1061_seed_10635_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 16:34:57.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:57.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:16.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_15/index_1061_seed_10636_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 16:36:39.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:36:40.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 16:48:15.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:24.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_15/index_2224_seed_3711_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 16:49:40.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:40.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:50:54.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_15/index_2224_seed_3712_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 16:51:16.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 17:02:55.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_15/index_4225_seed_7755_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 17:04:20.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:20.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:33.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_15/index_4225_seed_7756_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:05:57.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 17:17:35.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:45.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_15/index_4307_seed_20586_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 17:19:03.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:19:03.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:15.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_15/index_4307_seed_20587_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:20:38.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 17:32:10.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_15/index_6777_seed_31466_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 17:33:35.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:35.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:00.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_15/index_6777_seed_31467_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:35:26.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 17:46:54.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_15/index_13_seed_12269_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 17:48:18.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:18.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:49:38.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_15/index_13_seed_12270_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 17:50:01.302 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 17:50:02.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 18:01:35.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:44.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_15/index_7480_seed_1368_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 18:03:01.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:03:01.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:17.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_15/index_7480_seed_1369_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 18:04:39.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:04:40.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 18:16:18.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_15/index_1244_seed_26397_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 18:17:41.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:00.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_15/index_1244_seed_26398_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:19:32.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 18:31:18.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_15/index_4134_seed_169_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 18:32:44.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:44.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:14.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_15/index_4134_seed_170_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:34:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:17.786 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 18:47:24.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_15/index_3595_seed_23809_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 18:48:54.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:54.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:12.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_15/index_3595_seed_23810_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 18:50:39.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.535 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.142 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 19:02:13.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:26.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_15/index_9034_seed_6129_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 19:03:44.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:44.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:01.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_15/index_9034_seed_6130_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:05:34.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 19:17:12.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:22.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_15/index_392_seed_19608_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 19:18:41.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:41.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:19:59.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_15/index_392_seed_19609_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:20:26.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 19:31:55.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_15/index_9582_seed_26225_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 19:33:20.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:20.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:34:55.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_15/index_9582_seed_26226_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:35:24.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 19:47:03.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_15/index_8006_seed_4927_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 19:48:30.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:30.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:49:48.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_15/index_8006_seed_4928_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 19:50:11.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.986 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 20:01:46.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:57.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_15/index_5442_seed_491_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 20:03:15.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:03:15.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_15/index_5442_seed_492_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 20:04:55.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:04:56.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 20:16:29.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:41.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_15/index_2536_seed_13024_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 20:17:56.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:56.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:19.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_15/index_2536_seed_13025_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:19:47.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 20:31:20.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:29.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_15/index_6670_seed_15308_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 20:32:44.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:44.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:00.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_15/index_6670_seed_15309_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:34:24.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:02.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 20:46:03.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_15/index_3499_seed_27913_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 20:47:28.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:28.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:48:49.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_15/index_3499_seed_27914_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 20:49:15.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 21:00:47.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:48.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_15/index_3145_seed_27106_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 21:02:10.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:02:10.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:31.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_15/index_3145_seed_27107_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:03:58.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.128 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 21:17:31.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:41.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_15/index_9714_seed_13253_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 21:18:58.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:58.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:12.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_15/index_9714_seed_13254_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:20:38.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 21:32:09.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:49:17.040 | INFO | __main__:__init__:939 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-05-30 21:49:17.041 | INFO | __main__:__init__:940 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-05-30 21:49:17.722 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:49:17.722 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 21:55:22.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:48.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_15/index_717_seed_14730_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 21:57:04.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:57:04.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:17.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_15/index_717_seed_14731_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 21:58:52.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 21:59:08.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.883 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.941 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 62, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.387 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 22:10:52.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:53.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:12:00.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_15/index_9841_seed_6830_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 22:12:15.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:15.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:13:35.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_15/index_9841_seed_6831_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:14:03.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 22:25:33.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:42.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_15/index_6303_seed_21750_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 22:26:55.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:55.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:13.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_15/index_6303_seed_21751_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:28:37.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 22:40:03.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:12.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_15/index_6605_seed_20968_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 22:41:26.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:26.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:42:45.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_15/index_6605_seed_20969_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:43:10.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 22:54:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_15/index_2454_seed_12211_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 22:56:03.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:56:03.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:25.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_15/index_2454_seed_12212_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 22:57:49.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 22:57:50.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 23:09:20.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:27.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_15/index_8510_seed_13058_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 23:10:44.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:44.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:10.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_15/index_8510_seed_13059_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:12:36.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 23:24:05.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:14.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_15/index_4287_seed_30260_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 23:25:29.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:29.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:26:47.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_15/index_4287_seed_30261_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:27:14.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 23:38:46.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:55.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_15/index_8181_seed_26292_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 23:40:13.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:40:13.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:25.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_15/index_8181_seed_26293_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:41:48.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-30 23:53:10.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:18.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_15/index_9102_seed_26354_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-30 23:54:32.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:32.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:55:52.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_15/index_9102_seed_26355_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-30 23:56:15.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.209 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 00:07:41.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_15/index_4081_seed_18322_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 00:09:13.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:09:13.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:24.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_15/index_4081_seed_18323_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:10:46.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:53.789 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 00:22:59.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_15/index_7559_seed_13217_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 00:24:22.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:22.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:25:40.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_15/index_7559_seed_13218_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:26:05.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 00:37:30.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_15/index_2609_seed_17043_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 00:38:53.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:53.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:18.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_15/index_2609_seed_17044_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:40:46.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 00:52:07.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_15/index_1478_seed_28855_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 00:53:30.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:30.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:54:48.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_15/index_1478_seed_28856_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 00:55:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 01:06:44.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_15/index_3403_seed_21541_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 01:08:08.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:08:08.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:26.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_15/index_3403_seed_21542_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:09:52.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 01:21:16.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:17.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_15/index_626_seed_13193_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 01:22:38.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:23:55.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_15/index_626_seed_13194_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:24:20.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 01:35:36.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_15/index_9178_seed_3521_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 01:36:56.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:56.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:15.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_15/index_9178_seed_3522_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:38:38.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.35s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 01:50:03.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_15/index_7279_seed_10954_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 01:51:29.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:29.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:52:41.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_15/index_7279_seed_10955_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 01:53:07.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 02:04:34.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_15/index_2137_seed_4454_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 02:05:59.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:06:00.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:13.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_15/index_2137_seed_4455_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:07:39.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 62, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-05-31 02:19:01.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_15/index_1268_seed_19666_dp_31_rank_62_sample_2_slot_0.mp4 +2026-05-31 02:20:23.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:23.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:21:38.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_15/index_1268_seed_19667_dp_31_rank_62_sample_3_slot_1.mp4 +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-05-31 02:22:02.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:01:39.184 | INFO | __main__:__init__:946 - [RewardDebug] reward_weights(raw): {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0} +2026-06-25 18:01:39.186 | INFO | __main__:__init__:947 - [RewardDebug] reward_config(effective): {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'} +2026-06-25 18:01:39.845 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:01:39.907 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 18:04:09.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:31.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_15/index_717_seed_14730_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 18:05:47.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:47.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:06:56.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_15/index_717_seed_14731_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:07:32.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.293 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 62, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 18:18:41.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:47.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_15/index_9841_seed_6830_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 18:20:03.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:20:03.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:12.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000001/group_15/index_9841_seed_6831_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:21:35.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 18:32:26.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:32.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_15/index_6303_seed_21750_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 18:33:47.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:47.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:34:54.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000002/group_15/index_6303_seed_21751_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:35:15.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 18:46:08.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_15/index_6605_seed_20968_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 18:47:28.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:28.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:48:38.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000003/group_15/index_6605_seed_20969_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 18:49:04.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 19:00:02.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:09.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_15/index_2454_seed_12211_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 19:01:26.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:26.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:34.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000004/group_15/index_2454_seed_12212_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:02:57.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 19:13:49.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_15/index_8510_seed_13058_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 19:15:10.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:15:10.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:20.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000005/group_15/index_8510_seed_13059_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:16:42.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 19:27:32.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_15/index_4287_seed_30260_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 19:28:52.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:52.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:03.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000006/group_15/index_4287_seed_30261_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:30:24.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 19:41:17.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_15/index_8181_seed_26292_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 19:42:41.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:41.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:43:50.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000007/group_15/index_8181_seed_26293_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:44:15.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:08.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 19:55:08.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_15/index_9102_seed_26354_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 19:56:29.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:29.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:57:41.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000008/group_15/index_9102_seed_26355_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 19:58:03.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 20:09:02.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_15/index_4081_seed_18322_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 20:10:30.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:30.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:35.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000009/group_15/index_4081_seed_18323_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:12:00.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:30.009 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 20:23:32.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:40.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_15/index_7559_seed_13217_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 20:24:53.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:53.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:05.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000010/group_15/index_7559_seed_13218_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:26:26.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 20:37:18.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_15/index_2609_seed_17043_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 20:38:39.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:39.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:39:48.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000011/group_15/index_2609_seed_17044_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:40:09.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 20:51:03.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:09.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_15/index_1478_seed_28855_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 20:52:25.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:25.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:36.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000012/group_15/index_1478_seed_28856_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 20:53:57.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 21:04:50.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_15/index_3403_seed_21541_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 21:06:18.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:06:18.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:23.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000013/group_15/index_3403_seed_21542_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:07:49.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 21:18:43.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:44.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:49.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_15/index_626_seed_13193_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 21:20:07.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:20:07.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:17.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000014/group_15/index_626_seed_13194_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:21:39.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 21:32:31.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:37.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_15/index_9178_seed_3521_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 21:33:50.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:50.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:01.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000015/group_15/index_9178_seed_3522_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:35:21.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 21:46:16.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:22.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_15/index_7279_seed_10954_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 21:47:42.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:42.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:48:48.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000016/group_15/index_7279_seed_10955_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 21:49:10.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.238 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 22:00:05.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_15/index_2137_seed_4454_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 22:01:30.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:30.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:37.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000017/group_15/index_2137_seed_4455_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:02:58.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 22:13:50.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_15/index_1268_seed_19666_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 22:15:12.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:15:12.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:22.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000018/group_15/index_1268_seed_19667_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:16:45.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 22:27:39.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:45.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_15/index_2828_seed_2238_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 22:29:00.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:29:00.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:11.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000019/group_15/index_2828_seed_2239_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:30:32.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:57.965 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 22:42:00.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:07.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_15/index_7943_seed_2338_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 22:43:21.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:21.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:31.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000020/group_15/index_7943_seed_2339_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 22:44:51.321 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:44:52.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 22:55:45.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:51.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_15/index_324_seed_23323_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 22:57:04.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:57:04.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:17.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000021/group_15/index_324_seed_23324_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 22:58:38.406 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 22:58:39.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 23:09:30.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_15/index_3082_seed_14010_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 23:10:55.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:55.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:01.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000022/group_15/index_3082_seed_14011_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:12:22.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 23:23:15.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_15/index_7405_seed_7317_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 23:24:40.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:40.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:25:47.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000023/group_15/index_7405_seed_7318_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:26:11.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 23:37:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:13.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_15/index_2779_seed_26820_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 23:38:29.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:29.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:39:38.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000024/group_15/index_2779_seed_26821_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:40:02.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-25 23:51:52.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:58.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_15/index_5540_seed_8983_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-25 23:53:14.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:53:14.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:23.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000025/group_15/index_5540_seed_8984_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-25 23:54:43.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 00:05:36.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:42.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_15/index_1868_seed_8463_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 00:07:00.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:07:00.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:07.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000026/group_15/index_1868_seed_8464_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:08:28.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 00:19:21.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_15/index_921_seed_4158_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 00:20:39.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:39.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:21:50.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000027/group_15/index_921_seed_4159_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:22:11.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 00:33:07.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:13.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_15/index_9047_seed_7619_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 00:34:31.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:31.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:35:37.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000028/group_15/index_9047_seed_7620_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:36:04.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 00:47:03.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:10.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_15/index_3812_seed_19120_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 00:48:28.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:28.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:38.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000029/group_15/index_3812_seed_19121_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 00:50:00.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:28.914 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 01:01:30.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:31.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:38.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_15/index_144_seed_13716_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 01:02:54.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:54.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:04.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000030/group_15/index_144_seed_13717_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:04:25.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 01:15:18.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:24.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_15/index_6049_seed_13910_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 01:16:36.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:36.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:17:50.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000031/group_15/index_6049_seed_13911_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:18:11.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 01:29:08.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:14.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_15/index_9215_seed_1221_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 01:30:29.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:29.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:31:40.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000032/group_15/index_9215_seed_1222_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:32:03.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 01:42:58.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:04.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_15/index_3370_seed_27564_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 01:44:21.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:21.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:29.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000033/group_15/index_3370_seed_27565_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:45:50.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 01:56:44.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:50.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_15/index_9829_seed_23998_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 01:58:09.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:58:09.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:18.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000034/group_15/index_9829_seed_23999_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 01:59:40.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 02:10:31.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:37.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_15/index_3900_seed_28113_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 02:11:50.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:50.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:03.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000035/group_15/index_3900_seed_28114_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 02:13:23.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:13:25.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 02:24:16.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:22.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_15/index_9770_seed_8957_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 02:25:38.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:38.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:26:46.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000036/group_15/index_9770_seed_8958_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:27:11.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 02:38:08.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:14.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_15/index_2632_seed_26757_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 02:39:28.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:28.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:37.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000037/group_15/index_2632_seed_26758_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:41:00.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 02:51:53.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:54.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_15/index_2354_seed_7979_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 02:53:18.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:53:18.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:25.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000038/group_15/index_2354_seed_7980_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 02:54:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 03:05:40.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:41.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_15/index_1020_seed_28649_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 03:07:08.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:07:08.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:15.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000039/group_15/index_1020_seed_28650_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:08:38.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.686 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 03:20:06.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:14.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_15/index_5189_seed_24901_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 03:21:27.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:27.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:39.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000040/group_15/index_5189_seed_24902_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:22:59.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 03:33:52.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_15/index_2593_seed_24179_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 03:35:16.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:35:16.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:24.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000041/group_15/index_2593_seed_24180_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:36:46.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 03:47:40.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:46.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_15/index_655_seed_4248_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 03:48:59.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:59.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:10.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000042/group_15/index_655_seed_4249_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 03:50:31.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 04:01:21.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:27.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_15/index_124_seed_15797_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 04:02:42.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:42.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:03:52.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000043/group_15/index_124_seed_15798_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:04:12.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 04:15:09.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_15/index_1789_seed_31677_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 04:16:29.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:29.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:17:42.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000044/group_15/index_1789_seed_31678_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:18:05.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 04:29:03.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:10.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_15/index_5125_seed_1906_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 04:30:24.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:24.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:33.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000045/group_15/index_5125_seed_1907_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:31:54.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 04:42:46.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:52.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_15/index_852_seed_16014_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 04:44:11.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:44:11.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:16.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000046/group_15/index_852_seed_16015_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:45:35.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 04:56:27.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:34.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_15/index_5634_seed_4597_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 04:57:46.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:46.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:58:58.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000047/group_15/index_5634_seed_4598_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 04:59:19.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 05:10:10.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_15/index_3269_seed_28363_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 05:11:34.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:34.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:12:42.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000048/group_15/index_3269_seed_28364_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:13:04.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 05:24:03.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:09.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_15/index_1674_seed_30701_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 05:25:26.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:26.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:35.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000049/group_15/index_1674_seed_30702_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:26:58.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.127 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 05:39:29.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_15/index_4495_seed_30381_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 05:40:49.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:49.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:00.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000050/group_15/index_4495_seed_30382_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:42:20.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 05:53:10.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:16.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_15/index_3575_seed_28158_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 05:54:32.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:32.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:39.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000051/group_15/index_3575_seed_28159_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 05:56:00.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 06:06:52.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:58.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_15/index_7043_seed_8519_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 06:08:13.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:08:13.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:22.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000052/group_15/index_7043_seed_8520_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:09:44.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 06:20:40.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_15/index_9111_seed_21779_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 06:22:03.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:22:03.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:09.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000053/group_15/index_9111_seed_21780_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:23:31.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 06:34:21.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:22.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_15/index_1012_seed_7920_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 06:35:42.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:42.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:36:51.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000054/group_15/index_1012_seed_7921_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:37:11.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 06:48:09.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:15.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_15/index_3487_seed_14021_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 06:49:29.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:29.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:50:44.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000055/group_15/index_3487_seed_14022_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 06:51:03.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 07:02:03.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:09.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_15/index_4599_seed_25672_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 07:03:27.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:27.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000056/group_15/index_4599_seed_25673_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:04:55.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 07:15:48.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:54.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_15/index_6060_seed_15367_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 07:17:11.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:17:11.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:21.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000057/group_15/index_6060_seed_15368_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:18:41.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 07:29:32.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_15/index_1449_seed_23327_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 07:30:59.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:59.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000058/group_15/index_1449_seed_23328_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:32:31.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 07:43:21.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:27.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_15/index_4396_seed_27276_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 07:44:41.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:41.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:45:53.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000059/group_15/index_4396_seed_27277_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 07:46:15.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.954 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 07:57:46.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:53.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_15/index_3365_seed_5979_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 07:59:10.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:59:10.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:18.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000060/group_15/index_3365_seed_5980_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:00:40.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 08:11:33.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:39.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_15/index_1672_seed_23932_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 08:12:55.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:55.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:03.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000061/group_15/index_1672_seed_23933_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:14:26.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.464 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 08:25:18.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:24.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_15/index_6641_seed_21718_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 08:26:42.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:42.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:27:48.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000062/group_15/index_6641_seed_21719_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:28:08.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.658 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 08:39:03.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:09.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_15/index_3330_seed_7892_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 08:40:24.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:24.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:36.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000063/group_15/index_3330_seed_7893_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:41:57.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 08:52:50.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_15/index_3776_seed_26267_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 08:54:13.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:54:13.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:22.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000064/group_15/index_3776_seed_26268_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 08:55:45.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 09:06:39.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:45.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_15/index_6735_seed_18147_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 09:08:02.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:08:02.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:11.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000065/group_15/index_6735_seed_18148_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:09:33.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.953 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 09:20:22.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:28.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_15/index_1842_seed_9718_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 09:21:43.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:43.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:22:52.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000066/group_15/index_1842_seed_9719_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:23:12.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.693 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 09:34:09.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:15.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_15/index_3450_seed_7520_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 09:35:33.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:33.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:36:42.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000067/group_15/index_3450_seed_7521_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:37:04.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 09:47:56.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:03.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_15/index_779_seed_16678_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 09:49:19.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:19.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:29.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000068/group_15/index_779_seed_16679_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 09:50:51.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 10:01:45.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:51.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_15/index_5385_seed_11080_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 10:03:04.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:03:04.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:15.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000069/group_15/index_5385_seed_11081_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 10:04:34.031 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:04:35.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.942 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.945 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 10:16:01.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:09.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_15/index_3144_seed_24318_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 10:17:25.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:25.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:34.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000070/group_15/index_3144_seed_24319_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 10:18:58.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:18:59.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 10:29:52.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:58.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_15/index_1530_seed_11221_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 10:31:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:31:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:23.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000071/group_15/index_1530_seed_11222_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:32:44.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 10:43:37.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:43.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_15/index_151_seed_22664_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 10:44:57.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:57.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:08.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000072/group_15/index_151_seed_22665_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 10:46:29.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 10:57:21.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:27.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_15/index_9888_seed_14162_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 10:58:43.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:43.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:59:56.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000073/group_15/index_9888_seed_14163_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:00:18.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 11:11:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:09.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_15/index_713_seed_2943_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 11:12:28.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:28.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:39.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000074/group_15/index_713_seed_2944_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:13:59.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.580 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.293 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 11:26:08.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:14.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_15/index_8335_seed_7456_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 11:27:28.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:28.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:28:41.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000075/group_15/index_8335_seed_7457_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:29:03.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.252 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 11:39:55.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:01.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_15/index_8616_seed_20909_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 11:41:17.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:17.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:25.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000076/group_15/index_8616_seed_20910_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:42:44.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 11:53:37.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_15/index_8954_seed_12636_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 11:54:57.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:57.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:07.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000077/group_15/index_8954_seed_12637_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 11:56:31.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 12:07:21.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:27.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_15/index_4060_seed_5020_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 12:08:45.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:45.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:09:51.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000078/group_15/index_4060_seed_5021_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:10:12.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 12:21:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:07.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:13.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_15/index_2990_seed_13634_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 12:22:27.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:27.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:39.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000079/group_15/index_2990_seed_13635_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 12:23:59.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:24:00.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.819 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 12:35:30.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:39.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_15/index_3123_seed_443_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 12:36:53.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:53.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:04.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000080/group_15/index_3123_seed_444_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:38:25.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 12:49:16.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:22.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_15/index_1061_seed_10635_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 12:50:36.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:36.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:51:48.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000081/group_15/index_1061_seed_10636_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 12:52:08.745 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 12:52:09.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.810 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 13:03:03.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:09.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_15/index_2224_seed_3711_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 13:04:22.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:22.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:33.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000082/group_15/index_2224_seed_3712_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:05:53.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 13:16:44.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_15/index_4225_seed_7755_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 13:18:08.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:18:08.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:17.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000083/group_15/index_4225_seed_7756_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:19:41.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 13:30:31.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_15/index_4307_seed_20586_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 13:31:53.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:53.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:05.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000084/group_15/index_4307_seed_20587_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:33:25.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 13:44:15.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:21.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_15/index_6777_seed_31466_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 13:45:37.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:37.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:46:49.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000085/group_15/index_6777_seed_31467_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 13:47:12.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 13:47:13.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 13:58:07.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:13.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_15/index_13_seed_12269_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 13:59:28.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:29.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:36.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000086/group_15/index_13_seed_12270_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:00:57.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 14:11:49.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:55.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_15/index_7480_seed_1368_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 14:13:11.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:13:11.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:18.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000087/group_15/index_7480_seed_1369_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:14:39.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 14:25:28.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:34.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_15/index_1244_seed_26397_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 14:26:48.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:48.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:02.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000088/group_15/index_1244_seed_26398_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:28:25.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 14:39:15.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_15/index_4134_seed_169_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 14:40:35.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:35.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:41:50.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000089/group_15/index_4134_seed_170_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:42:13.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.334 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 14:53:45.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:53.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_15/index_3595_seed_23809_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 14:55:07.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:55:07.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:16.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000090/group_15/index_3595_seed_23810_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 14:56:36.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 15:07:27.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:33.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_15/index_9034_seed_6129_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 15:08:50.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:50.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:09:56.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000091/group_15/index_9034_seed_6130_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:10:19.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.763 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 15:21:11.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:17.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_15/index_392_seed_19608_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 15:22:34.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:34.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:23:41.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000092/group_15/index_392_seed_19609_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:24:02.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 15:34:56.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_15/index_9582_seed_26225_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 15:36:16.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:16.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:30.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000093/group_15/index_9582_seed_26226_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:37:50.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.565 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 15:48:45.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:51.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_15/index_8006_seed_4927_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 15:50:08.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:50:08.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:15.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000094/group_15/index_8006_seed_4928_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 15:51:35.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 16:02:26.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_15/index_5442_seed_491_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 16:03:51.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:51.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:04:57.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000095/group_15/index_5442_seed_492_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:05:18.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 16:16:06.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:07.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_15/index_2536_seed_13024_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 16:17:28.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:28.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:18:42.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000096/group_15/index_2536_seed_13025_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:19:04.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.424 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 16:29:54.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_15/index_6670_seed_15308_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 16:31:14.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:14.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:24.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000097/group_15/index_6670_seed_15309_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 16:32:44.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:32:45.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 16:43:38.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_15/index_3499_seed_27913_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 16:45:00.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:45:00.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:08.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000098/group_15/index_3499_seed_27914_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 16:46:31.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 16:57:20.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:21.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_15/index_3145_seed_27106_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 16:58:44.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:44.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:59:54.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000099/group_15/index_3145_seed_27107_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:00:16.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.057 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 17:12:52.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:14:00.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_15/index_9714_seed_13253_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 17:14:17.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:14:17.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:26.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000100/group_15/index_9714_seed_13254_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 17:15:47.056 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:15:48.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 17:26:39.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:45.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_15/index_4454_seed_26990_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 17:27:59.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:59.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:09.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000101/group_15/index_4454_seed_26991_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:29:30.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 17:40:20.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:26.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_15/index_2039_seed_5401_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 17:41:39.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:39.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:42:54.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000102/group_15/index_2039_seed_5402_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:43:14.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.959 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 17:54:06.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:12.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_15/index_6637_seed_23244_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 17:55:29.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:29.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:38.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000103/group_15/index_6637_seed_23245_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 17:56:59.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.657 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 18:07:49.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:55.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_15/index_4835_seed_11891_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 18:09:11.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:09:11.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:21.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000104/group_15/index_4835_seed_11892_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:10:46.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 18:21:38.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:44.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_15/index_167_seed_31814_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 18:22:57.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:57.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:12.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000105/group_15/index_167_seed_31815_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 18:24:36.191 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:24:37.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 18:35:27.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_15/index_9703_seed_26620_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 18:36:50.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:50.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:03.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000106/group_15/index_9703_seed_26621_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:38:24.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.305 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 18:49:15.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_15/index_2528_seed_4649_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 18:50:36.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:36.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:51:45.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000107/group_15/index_2528_seed_4650_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 18:52:05.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.77s (1.88min) +2026-06-26 18:53:58.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 19:03:02.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:08.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_15/index_2345_seed_14118_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 19:04:22.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:22.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:35.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000108/group_15/index_2345_seed_14119_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:05:55.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 19:16:46.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:52.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_15/index_3954_seed_30015_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 19:18:05.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:18:05.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:20.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000109/group_15/index_3954_seed_30016_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:19:40.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.001 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.525 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 19:31:08.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:17.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_15/index_5129_seed_11903_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 19:32:37.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:37.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:33:43.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000110/group_15/index_5129_seed_11904_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:34:07.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 19:45:03.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:10.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_15/index_4843_seed_32836_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 19:46:25.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:25.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:32.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000111/group_15/index_4843_seed_32837_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 19:47:52.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 19:58:45.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:52.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_15/index_1950_seed_12595_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 20:00:06.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:00:06.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:17.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000112/group_15/index_1950_seed_12596_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:01:37.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.572 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 20:12:28.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:34.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_15/index_2485_seed_727_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 20:13:48.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:48.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:14:59.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000113/group_15/index_2485_seed_728_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:15:22.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 20:26:14.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_15/index_2502_seed_12081_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 20:27:40.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:40.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:28:48.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000114/group_15/index_2502_seed_12082_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:29:12.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 20:40:07.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_15/index_7970_seed_15877_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 20:41:27.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:27.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:42:39.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000115/group_15/index_7970_seed_15878_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:43:01.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 20:53:52.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:53.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:59.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_15/index_2226_seed_28014_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 20:55:15.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:55:15.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:25.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000116/group_15/index_2226_seed_28015_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 20:56:54.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 21:07:44.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:50.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_15/index_4916_seed_31054_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 21:09:04.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:09:04.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:18.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000117/group_15/index_4916_seed_31055_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:10:42.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 21:21:31.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:37.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_15/index_4285_seed_30681_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 21:22:56.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:56.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:02.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000118/group_15/index_4285_seed_30682_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:24:26.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.02s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 21:35:16.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:23.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_15/index_3720_seed_30584_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 21:36:40.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:40.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:37:49.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000119/group_15/index_3720_seed_30585_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:38:11.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.820 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 21:49:43.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:43.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:51.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_15/index_4651_seed_6835_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 21:51:06.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:51:06.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:15.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000120/group_15/index_4651_seed_6836_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 21:52:38.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 22:03:26.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:27.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_15/index_2158_seed_11196_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 22:04:49.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:49.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:05:56.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000121/group_15/index_2158_seed_11197_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:06:18.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 22:17:07.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:13.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_15/index_4898_seed_25141_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 22:18:32.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:32.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:37.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000122/group_15/index_4898_seed_25142_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 22:19:58.738 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:19:59.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 22:30:49.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:55.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_15/index_9958_seed_5144_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 22:32:09.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:32:09.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:20.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000123/group_15/index_9958_seed_5145_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:33:43.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 22:44:32.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:38.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_15/index_3491_seed_8973_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 22:45:53.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:53.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:04.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000124/group_15/index_3491_seed_8974_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 22:47:24.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 22:59:16.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_15/index_1733_seed_23495_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 23:00:35.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:35.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:01:48.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000125/group_15/index_1733_seed_23496_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:02:09.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 23:13:02.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:09.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_15/index_9549_seed_27772_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 23:14:25.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:25.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:32.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000126/group_15/index_9549_seed_27773_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:15:53.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 23:26:44.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:50.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_15/index_4063_seed_636_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 23:28:06.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:28:06.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:16.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000127/group_15/index_4063_seed_637_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:29:39.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 23:40:27.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:33.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_15/index_5825_seed_21367_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 23:41:48.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:48.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:01.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000128/group_15/index_5825_seed_21368_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:43:26.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.672 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-26 23:54:16.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:22.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_15/index_1172_seed_32716_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-26 23:55:41.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:41.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:56:47.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000129/group_15/index_1172_seed_32717_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-26 23:57:09.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.234 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 00:08:40.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:48.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_15/index_2985_seed_3312_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 00:10:06.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:10:06.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:12.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000130/group_15/index_2985_seed_3313_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:11:32.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 00:22:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:26.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_15/index_5169_seed_26505_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 00:23:45.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:45.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:24:54.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000131/group_15/index_5169_seed_26506_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:25:15.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 00:36:06.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:12.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_15/index_1541_seed_19531_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 00:37:29.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:29.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:38.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000132/group_15/index_1541_seed_19532_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:38:59.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.007 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 00:49:49.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:55.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_15/index_6468_seed_25658_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 00:51:09.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:51:09.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:19.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000133/group_15/index_6468_seed_25659_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 00:52:45.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 01:03:35.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_15/index_245_seed_13148_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 01:04:54.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:54.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000134/group_15/index_245_seed_13149_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:06:28.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 01:17:20.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:20.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_15/index_7788_seed_30701_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 01:18:41.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:41.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:19:52.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000135/group_15/index_7788_seed_30702_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:20:15.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 01:31:07.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_15/index_801_seed_16663_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 01:32:27.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:27.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:39.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000136/group_15/index_801_seed_16664_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:33:58.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.927 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 01:44:47.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_15/index_1568_seed_13152_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 01:46:07.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:46:07.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:19.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000137/group_15/index_1568_seed_13153_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 01:47:40.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.710 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 01:58:27.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:33.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_15/index_7713_seed_25772_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 01:59:47.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:47.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:01.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000138/group_15/index_7713_seed_25773_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 02:01:22.441 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:01:23.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 02:12:14.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:15.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_15/index_8393_seed_8606_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 02:13:36.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:36.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:14:47.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000139/group_15/index_8393_seed_8607_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:15:09.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.397 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 02:26:39.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:47.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_15/index_8641_seed_26036_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 02:28:05.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:28:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:13.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000140/group_15/index_8641_seed_26037_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:29:33.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 02:40:23.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:29.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_15/index_9742_seed_18008_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 02:41:46.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:46.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:00.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000141/group_15/index_9742_seed_18009_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:43:25.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 02:54:14.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:20.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_15/index_9869_seed_7145_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 02:55:38.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:38.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:56:47.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000142/group_15/index_9869_seed_7146_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 02:57:09.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 03:08:01.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:02.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:08.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_15/index_1701_seed_32662_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 03:09:23.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:23.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:31.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000143/group_15/index_1701_seed_32663_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:10:50.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 03:21:41.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:47.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_15/index_5393_seed_353_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 03:23:01.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:23:01.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:15.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000144/group_15/index_5393_seed_354_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:24:38.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 03:35:26.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:32.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_15/index_3186_seed_32783_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 03:36:50.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:50.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:02.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000145/group_15/index_3186_seed_32784_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:38:25.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 03:49:14.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:20.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_15/index_4269_seed_6993_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 03:50:33.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:33.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:51:45.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000146/group_15/index_4269_seed_6994_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 03:52:08.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 04:03:03.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:09.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_15/index_1624_seed_27787_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 04:04:23.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:23.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:33.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000147/group_15/index_1624_seed_27788_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:05:55.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 04:16:48.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_15/index_4012_seed_27660_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 04:18:14.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:18:14.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:20.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000148/group_15/index_4012_seed_27661_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:19:42.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 04:30:31.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_15/index_382_seed_32370_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 04:31:51.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:51.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:01.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000149/group_15/index_382_seed_32371_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:33:22.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.676 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 04:45:58.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_15/index_9891_seed_17563_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 04:47:24.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:24.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:31.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000150/group_15/index_9891_seed_17564_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 04:48:56.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 04:59:46.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:53.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_15/index_3713_seed_20497_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 05:01:10.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:01:10.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:17.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000151/group_15/index_3713_seed_20498_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:02:40.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 05:13:28.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:33.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_15/index_30_seed_20840_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 05:14:52.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:52.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:15:57.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000152/group_15/index_30_seed_20841_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:16:17.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 05:27:08.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:08.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_15/index_9977_seed_20497_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 05:28:29.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:29.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:29:40.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000153/group_15/index_9977_seed_20498_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:30:01.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 05:40:50.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:56.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_15/index_4833_seed_4878_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 05:42:13.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:42:13.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:19.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000154/group_15/index_4833_seed_4879_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:43:44.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 05:54:35.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:41.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_15/index_4308_seed_1151_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 05:55:58.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:58.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:07.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000155/group_15/index_4308_seed_1152_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 05:57:30.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 06:08:19.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_15/index_751_seed_33219_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 06:09:42.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:42.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:10:50.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000156/group_15/index_751_seed_33220_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:11:14.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 06:22:08.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_15/index_1041_seed_18266_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 06:23:28.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:28.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:39.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000157/group_15/index_1041_seed_18267_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 06:24:58.738 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:24:59.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 06:35:50.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:56.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_15/index_8503_seed_31236_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 06:37:11.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:37:11.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:20.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000158/group_15/index_8503_seed_31237_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:38:42.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 06:49:30.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:36.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_15/index_6547_seed_11049_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 06:50:51.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:51.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:02.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000159/group_15/index_6547_seed_11050_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 06:52:21.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.915 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 07:03:49.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_15/index_95_seed_21127_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 07:05:14.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:05:14.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:20.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000160/group_15/index_95_seed_21128_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:06:40.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 07:17:29.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:35.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_15/index_9303_seed_16775_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 07:18:49.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:49.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:00.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000161/group_15/index_9303_seed_16776_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 07:20:21.813 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:20:22.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.522 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 07:31:16.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:22.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_15/index_923_seed_21209_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 07:32:36.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:36.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:33:49.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000162/group_15/index_923_seed_21210_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:34:12.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 07:45:07.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_15/index_4723_seed_5744_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 07:46:32.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:32.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:47:38.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000163/group_15/index_4723_seed_5745_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 07:48:02.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 07:58:52.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_15/index_9191_seed_28620_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 08:00:12.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:00:12.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:23.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000164/group_15/index_9191_seed_28621_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:01:46.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 08:12:38.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_15/index_2372_seed_22700_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 08:14:05.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:14:05.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:10.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000165/group_15/index_2372_seed_22701_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:15:31.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.767 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 08:26:20.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:26.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_15/index_5994_seed_20347_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 08:27:40.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:40.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:28:51.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000166/group_15/index_5994_seed_20348_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:29:10.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 08:40:02.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:09.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_15/index_8592_seed_31852_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 08:41:23.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:23.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:35.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000167/group_15/index_8592_seed_31853_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:42:57.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 08:53:47.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:48.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:54.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_15/index_5480_seed_14262_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 08:55:08.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:55:08.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:23.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000168/group_15/index_5480_seed_14263_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 08:56:50.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 09:07:41.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:47.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_15/index_3430_seed_8855_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 09:09:05.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:09:05.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000169/group_15/index_3430_seed_8856_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:10:37.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.575 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:01.157 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 09:22:02.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:02.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:10.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_15/index_6828_seed_26238_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 09:23:24.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:24.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:34.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000170/group_15/index_6828_seed_26239_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:24:55.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 09:35:43.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_15/index_4072_seed_21440_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 09:37:08.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:37:08.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000171/group_15/index_4072_seed_21441_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:38:35.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 09:49:26.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_15/index_7440_seed_30991_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 09:50:47.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:47.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:51:57.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000172/group_15/index_7440_seed_30992_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 09:52:17.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.247 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 10:03:08.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_15/index_1687_seed_8096_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 10:04:30.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:30.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:39.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000173/group_15/index_1687_seed_8097_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 10:05:59.863 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:06:00.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 10:16:53.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_15/index_8366_seed_26946_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 10:18:18.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:18:18.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:24.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000174/group_15/index_8366_seed_26947_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:19:46.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 10:31:49.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_15/index_248_seed_11296_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 10:33:10.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:33:10.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:21.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000175/group_15/index_248_seed_11297_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:34:43.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 10:45:31.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_15/index_879_seed_9285_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 10:46:52.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:52.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:03.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000176/group_15/index_879_seed_9286_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 10:48:27.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 10:59:16.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_15/index_1004_seed_11708_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 11:00:41.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:41.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:01:49.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000177/group_15/index_1004_seed_11709_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:02:13.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 11:13:07.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_15/index_5318_seed_30711_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 11:14:29.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:29.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:35.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000178/group_15/index_5318_seed_30712_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:15:56.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 11:26:47.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:53.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_15/index_4163_seed_3956_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 11:28:09.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:28:09.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:21.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000179/group_15/index_4163_seed_3957_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:29:42.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.439 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 11:41:06.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_15/index_1646_seed_13687_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 11:42:27.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:27.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:38.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000180/group_15/index_1646_seed_13688_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 11:43:58.131 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:43:59.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 11:54:48.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_15/index_8613_seed_19087_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 11:56:10.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:56:10.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:19.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000181/group_15/index_8613_seed_19088_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 11:57:45.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 12:08:34.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_15/index_1164_seed_2595_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 12:09:54.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:54.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:05.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000182/group_15/index_1164_seed_2596_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:11:27.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 12:22:19.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:25.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_15/index_5140_seed_31639_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 12:23:39.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:39.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:24:50.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000183/group_15/index_5140_seed_31640_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:25:10.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 12:36:01.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_15/index_348_seed_8710_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 12:37:23.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:23.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:33.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000184/group_15/index_348_seed_8711_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 12:38:52.462 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:38:53.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 12:49:44.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:50.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_15/index_746_seed_31353_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 12:51:05.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:51:05.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:14.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000185/group_15/index_746_seed_31354_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 12:52:34.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 13:03:21.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:27.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_15/index_9952_seed_29616_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 13:04:42.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:42.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:05:51.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000186/group_15/index_9952_seed_29617_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:06:13.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 13:17:08.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_15/index_7000_seed_22357_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 13:18:29.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:29.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:37.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000187/group_15/index_7000_seed_22358_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:19:58.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.410 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.516 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.375 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 13:30:47.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_15/index_7354_seed_2738_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 13:32:06.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:32:06.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:21.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000188/group_15/index_7354_seed_2739_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:33:47.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 13:44:38.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_15/index_7414_seed_13934_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 13:45:59.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:59.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000189/group_15/index_7414_seed_13935_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 13:47:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.124 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 13:58:57.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:03.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_15/index_747_seed_31705_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 14:00:22.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:22.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:30.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000190/group_15/index_747_seed_31706_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:01:51.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 14:12:43.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:49.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_15/index_9207_seed_10994_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 14:14:04.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:14:04.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:12.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000191/group_15/index_9207_seed_10995_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 14:15:31.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:15:32.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 14:26:20.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_15/index_3449_seed_12833_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 14:27:45.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:28:50.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000192/group_15/index_3449_seed_12834_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:29:12.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:05.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 14:40:05.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_15/index_5768_seed_12018_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 14:41:30.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:30.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:36.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000193/group_15/index_5768_seed_12019_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:42:58.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 14:53:47.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_15/index_5696_seed_22180_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 14:55:07.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:55:07.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:18.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000194/group_15/index_5696_seed_22181_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 14:56:39.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 15:07:26.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_15/index_205_seed_26685_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 15:08:48.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:48.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:09:56.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000195/group_15/index_205_seed_26686_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:10:18.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 15:21:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:12.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_15/index_1729_seed_16819_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 15:22:30.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:30.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:37.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000196/group_15/index_1729_seed_16820_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:23:58.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 15:34:47.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:47.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_15/index_2651_seed_8499_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 15:36:08.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:36:08.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:20.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000197/group_15/index_2651_seed_8500_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:37:44.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.124 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 15:48:32.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_15/index_2675_seed_5525_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 15:49:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:52.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:03.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000198/group_15/index_2675_seed_5526_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 15:51:25.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.392 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 16:02:14.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:20.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_15/index_6462_seed_4129_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 16:03:34.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:34.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:04:45.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000199/group_15/index_6462_seed_4130_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:05:05.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.882 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 16:17:43.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_15/index_5583_seed_1668_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 16:19:05.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:19:05.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:17.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000200/group_15/index_5583_seed_1669_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:20:37.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 16:31:26.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_15/index_5239_seed_13144_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 16:32:45.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:45.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:33:58.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000201/group_15/index_5239_seed_13145_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:34:18.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 16:45:07.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_15/index_9000_seed_6422_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 16:46:29.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:29.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:39.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000202/group_15/index_9000_seed_6423_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 16:47:59.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 16:58:47.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_15/index_7655_seed_14523_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 17:00:07.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:00:07.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:20.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000203/group_15/index_7655_seed_14524_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:01:42.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 17:12:31.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:37.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_15/index_7787_seed_14563_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 17:13:51.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:51.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:02.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000204/group_15/index_7787_seed_14564_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:15:22.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 17:26:15.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_15/index_5765_seed_25619_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 17:27:38.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:38.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:28:46.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000205/group_15/index_5765_seed_25620_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 17:29:07.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:29:09.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 17:40:00.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:06.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_15/index_3407_seed_7243_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 17:41:21.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:21.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:34.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000206/group_15/index_3407_seed_7244_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:42:55.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 17:53:45.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:51.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_15/index_8446_seed_32874_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 17:55:07.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:55:07.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:15.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000207/group_15/index_8446_seed_32875_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 17:56:38.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 18:07:25.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:31.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_15/index_6910_seed_31528_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 18:08:50.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:50.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:01.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000208/group_15/index_6910_seed_31529_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:10:24.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 18:21:14.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:20.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_15/index_5254_seed_3017_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 18:22:35.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:35.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:23:45.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000209/group_15/index_5254_seed_3018_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:24:07.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.569 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.719 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 18:35:38.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:44.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_15/index_7694_seed_2617_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 18:36:57.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:57.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:09.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000210/group_15/index_7694_seed_2618_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:38:32.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 18:49:21.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_15/index_2359_seed_11926_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 18:50:40.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:40.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:51:55.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000211/group_15/index_2359_seed_11927_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 18:52:16.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 19:03:06.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:12.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_15/index_1044_seed_13376_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 19:04:26.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:26.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:36.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000212/group_15/index_1044_seed_13377_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:05:55.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.165 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 19:16:46.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_15/index_3535_seed_15037_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 19:18:06.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:18:06.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:21.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000213/group_15/index_3535_seed_15038_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:19:42.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 19:30:30.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:36.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_15/index_8079_seed_28718_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 19:31:52.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:52.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:04.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000214/group_15/index_8079_seed_28719_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:33:25.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.319 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 19:44:14.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_15/index_6553_seed_24396_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 19:45:34.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:34.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:46:44.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000215/group_15/index_6553_seed_24397_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 19:47:10.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 19:58:02.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:08.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_15/index_964_seed_35_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 19:59:24.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:24.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:33.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000216/group_15/index_964_seed_36_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:00:54.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 20:11:43.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_15/index_4048_seed_18575_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 20:13:11.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:13:11.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:16.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000217/group_15/index_4048_seed_18576_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:14:39.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 20:25:26.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:32.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_15/index_4699_seed_19714_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 20:26:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:48.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:27:59.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000218/group_15/index_4699_seed_19715_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 20:28:17.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:28:19.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.437 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 20:39:06.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_15/index_8581_seed_17415_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 20:40:27.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:27.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:41:38.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000219/group_15/index_8581_seed_17416_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 20:42:03.521 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:42:04.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.564 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 20:53:29.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:35.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_15/index_5093_seed_6922_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 20:54:50.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:50.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:55:59.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000220/group_15/index_5093_seed_6923_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 20:56:20.307 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 20:56:21.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 21:07:10.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:16.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_15/index_8620_seed_26911_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 21:08:31.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:31.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:09:41.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000221/group_15/index_8620_seed_26912_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:10:06.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 21:21:01.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_15/index_760_seed_28351_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 21:22:20.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:20.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:31.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000222/group_15/index_760_seed_28352_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:23:51.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 21:34:42.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:48.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_15/index_5475_seed_3544_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 21:36:02.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:36:02.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:13.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000223/group_15/index_5475_seed_3545_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 21:37:33.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:37:34.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.971 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 21:48:23.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:23.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:30.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_15/index_6628_seed_3980_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 21:49:48.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:48.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:50:55.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000224/group_15/index_6628_seed_3981_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 21:51:15.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 22:03:21.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:27.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_15/index_6400_seed_23621_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 22:04:44.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:44.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:05:54.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000225/group_15/index_6400_seed_23622_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 22:06:16.001 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:06:17.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 22:17:06.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:06.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:12.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_15/index_5597_seed_12700_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 22:18:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:30.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:37.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000226/group_15/index_5597_seed_12701_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:19:57.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 22:30:47.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:53.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_15/index_2030_seed_5797_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 22:32:11.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:32:11.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:21.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000227/group_15/index_2030_seed_5798_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:33:41.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 22:44:30.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:36.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_15/index_3873_seed_1443_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 22:45:50.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:50.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:00.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000228/group_15/index_3873_seed_1444_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 22:47:18.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 22:58:07.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_15/index_1477_seed_20634_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 22:59:31.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:31.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:00:38.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000229/group_15/index_1477_seed_20635_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:01:01.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.803 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 23:12:29.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_15/index_9294_seed_5779_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 23:13:57.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:57.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:02.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000230/group_15/index_9294_seed_5780_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:15:23.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 23:26:15.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_15/index_1660_seed_24098_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 23:27:36.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:36.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:28:45.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000231/group_15/index_1660_seed_24099_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 23:29:05.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:29:06.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 23:40:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:08.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_15/index_1566_seed_13572_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 23:41:24.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:24.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:42:36.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000232/group_15/index_1566_seed_13573_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:43:01.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-27 23:53:51.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:57.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_15/index_3010_seed_2859_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-27 23:55:11.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:55:11.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:22.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000233/group_15/index_3010_seed_2860_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-27 23:56:44.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 00:07:32.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:38.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_15/index_9266_seed_8392_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 00:08:56.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:56.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:08.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000234/group_15/index_9266_seed_8393_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:10:31.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 00:21:21.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:22.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_15/index_7735_seed_20933_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 00:22:44.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:44.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:23:51.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000235/group_15/index_7735_seed_20934_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:24:12.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 00:35:07.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:13.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_15/index_9401_seed_23142_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 00:36:26.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:26.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:37:45.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000236/group_15/index_9401_seed_23143_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:38:10.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 00:49:02.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:03.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:09.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_15/index_1025_seed_11694_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 00:50:22.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:22.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:36.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000237/group_15/index_1025_seed_11695_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 00:51:57.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 01:02:48.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:55.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_15/index_7366_seed_7659_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 01:04:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:04:09.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:21.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000238/group_15/index_7366_seed_7660_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:05:44.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 01:16:33.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:39.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_15/index_156_seed_18936_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 01:17:56.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:56.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:11.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000239/group_15/index_156_seed_18937_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:19:31.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.994 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 01:30:58.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_15/index_6965_seed_2820_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 01:32:19.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:19.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:30.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000240/group_15/index_6965_seed_2821_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:33:52.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 01:44:44.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:50.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_15/index_6889_seed_582_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 01:46:05.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:46:05.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:16.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000241/group_15/index_6889_seed_583_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 01:47:37.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 01:58:27.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:33.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_15/index_4345_seed_6903_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 01:59:46.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:46.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:00:59.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000242/group_15/index_4345_seed_6904_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 02:01:17.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:01:18.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.893 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 02:12:08.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:14.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_15/index_703_seed_21521_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 02:13:33.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:33.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:14:41.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000243/group_15/index_703_seed_21522_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:15:04.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 02:25:53.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_15/index_6707_seed_21319_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 02:27:20.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:27:20.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:25.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000244/group_15/index_6707_seed_21320_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:28:45.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 02:39:38.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:44.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_15/index_9306_seed_11925_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 02:41:01.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:41:01.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:08.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000245/group_15/index_9306_seed_11926_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:42:31.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 02:53:20.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:21.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_15/index_5647_seed_32903_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 02:54:41.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:41.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:55:51.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000246/group_15/index_5647_seed_32904_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 02:56:13.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 03:07:08.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_15/index_230_seed_17595_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 03:08:29.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:29.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:09:38.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000247/group_15/index_230_seed_17596_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 03:10:01.813 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:10:03.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 03:20:53.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:59.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_15/index_9407_seed_32664_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 03:22:16.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:22:16.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:26.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000248/group_15/index_9407_seed_32665_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:23:49.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 03:34:40.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:46.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_15/index_6193_seed_13619_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 03:36:03.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:36:03.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:12.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000249/group_15/index_6193_seed_13620_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:37:35.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.823 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 03:50:13.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_15/index_535_seed_4520_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 03:51:34.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:34.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:52:46.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000250/group_15/index_535_seed_4521_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 03:53:06.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 04:04:03.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:09.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_15/index_8703_seed_30212_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 04:05:25.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:25.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:33.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000251/group_15/index_8703_seed_30213_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:06:56.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 04:17:48.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_15/index_3546_seed_26926_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 04:19:13.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:19:13.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:20.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000252/group_15/index_3546_seed_26927_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:20:47.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 04:31:40.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:46.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_15/index_7580_seed_15800_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 04:33:04.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:33:04.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:09.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000253/group_15/index_7580_seed_15801_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:34:29.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 04:45:20.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:26.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_15/index_8941_seed_3323_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 04:46:41.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:41.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:47:50.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000254/group_15/index_8941_seed_3324_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 04:48:10.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 04:59:03.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:09.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_15/index_7605_seed_3790_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 05:00:23.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:23.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:32.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000255/group_15/index_7605_seed_3791_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:01:52.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 05:12:44.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:45.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:50.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_15/index_4021_seed_22001_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 05:14:03.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:14:03.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:15.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000256/group_15/index_4021_seed_22002_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:15:35.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 05:26:26.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:32.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_15/index_1971_seed_24480_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 05:27:48.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:48.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:28:56.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000257/group_15/index_1971_seed_24481_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:29:22.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.186 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 05:40:15.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:21.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_15/index_7990_seed_27743_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 05:41:33.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:33.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:42:47.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000258/group_15/index_7990_seed_27744_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 05:43:09.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:43:10.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 05:54:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:07.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_15/index_9309_seed_29934_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 05:55:29.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:29.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:38.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000259/group_15/index_9309_seed_29935_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 05:57:00.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.705 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.922 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 06:08:27.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:33.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_15/index_189_seed_16970_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 06:09:50.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:50.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:10:58.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000260/group_15/index_189_seed_16971_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:11:21.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 06:22:11.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:17.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_15/index_4608_seed_1716_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 06:23:30.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:30.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:24:42.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000261/group_15/index_4608_seed_1717_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 06:25:06.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:25:07.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 06:36:02.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_15/index_653_seed_23675_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 06:37:23.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:23.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:32.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000262/group_15/index_653_seed_23676_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:38:53.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:44.002 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 06:49:45.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:51.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_15/index_3287_seed_10821_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 06:51:03.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:51:03.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:16.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000263/group_15/index_3287_seed_10822_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 06:52:37.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:27.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 07:03:27.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:33.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_15/index_7037_seed_3419_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 07:04:48.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:48.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:05:59.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000264/group_15/index_7037_seed_3420_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:06:21.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 07:17:12.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_15/index_7410_seed_26518_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 07:18:37.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:37.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:19:42.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000265/group_15/index_7410_seed_26519_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:20:04.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 07:30:53.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:59.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_15/index_6823_seed_19106_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 07:32:14.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:32:14.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:26.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000266/group_15/index_6823_seed_19107_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:33:48.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 07:44:40.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:46.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_15/index_2619_seed_20508_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 07:46:06.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:46:06.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:12.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000267/group_15/index_2619_seed_20509_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 07:47:32.045 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 07:47:33.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 07:58:21.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_15/index_4055_seed_20587_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 07:59:42.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:42.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:00:52.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000268/group_15/index_4055_seed_20588_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:01:16.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 08:12:08.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:14.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_15/index_2858_seed_28356_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 08:13:30.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:30.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:14:39.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000269/group_15/index_2858_seed_28357_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 08:15:02.254 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:15:03.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.541 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 08:26:31.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:37.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_15/index_7221_seed_3512_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 08:27:56.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:56.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:02.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000270/group_15/index_7221_seed_3513_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:29:27.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 08:40:22.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:29.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_15/index_2479_seed_3899_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 08:41:43.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:43.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:42:53.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000271/group_15/index_2479_seed_3900_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:43:14.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 08:54:07.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_15/index_2735_seed_18841_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 08:55:26.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:26.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:38.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000272/group_15/index_2735_seed_18842_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 08:57:00.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 09:07:50.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_15/index_6034_seed_9539_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 09:09:13.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:09:13.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:21.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000273/group_15/index_6034_seed_9540_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:10:44.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 09:21:32.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:38.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_15/index_5023_seed_11410_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 09:22:55.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:55.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:04.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000274/group_15/index_5023_seed_11411_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:24:28.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.229 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 09:36:35.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_15/index_4061_seed_6940_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 09:37:57.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:57.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:05.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000275/group_15/index_4061_seed_6941_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:39:27.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 09:50:22.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:28.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_15/index_1324_seed_7244_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 09:51:42.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:42.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:52:55.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000276/group_15/index_1324_seed_7245_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 09:53:25.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 10:04:16.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:22.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_15/index_5933_seed_22464_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 10:05:42.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:42.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:06:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000277/group_15/index_5933_seed_22465_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:07:11.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 10:18:09.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:15.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_15/index_9473_seed_4363_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 10:19:31.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:31.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:20:40.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000278/group_15/index_9473_seed_4364_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:21:02.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.744 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 10:31:54.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:33:00.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_15/index_2793_seed_7198_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 10:33:15.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:33:15.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:26.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000279/group_15/index_2793_seed_7199_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:34:54.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.943 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.669 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 10:46:20.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:27.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_15/index_3412_seed_13594_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 10:47:41.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:41.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:48:51.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000280/group_15/index_3412_seed_13595_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 10:49:13.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.308 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 11:00:08.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_15/index_926_seed_25958_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 11:01:29.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:29.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:39.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000281/group_15/index_926_seed_25959_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:03:00.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.641 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 11:13:51.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_15/index_2795_seed_1375_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 11:15:12.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:15:12.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:21.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000282/group_15/index_2795_seed_1376_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 11:16:42.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:16:43.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 11:27:32.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:38.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_15/index_9786_seed_22261_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 11:28:53.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:53.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:02.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000283/group_15/index_9786_seed_22262_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:30:23.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.909 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 11:41:15.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:21.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_15/index_6856_seed_24511_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 11:42:35.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:35.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:43:47.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000284/group_15/index_6856_seed_24512_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:44:08.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 11:55:03.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:09.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_15/index_9785_seed_20190_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 11:56:26.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:26.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:34.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000285/group_15/index_9785_seed_20191_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 11:57:57.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 12:08:48.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:54.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_15/index_372_seed_3577_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 12:10:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:10:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:25.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000286/group_15/index_372_seed_3578_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:11:48.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 12:22:40.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:46.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_15/index_6078_seed_2578_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 12:24:04.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:24:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:13.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000287/group_15/index_6078_seed_2579_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:25:34.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 12:36:24.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:30.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_15/index_9892_seed_24618_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 12:37:45.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:45.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:38:55.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000288/group_15/index_9892_seed_24619_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:39:17.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 12:50:08.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_15/index_8881_seed_10697_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 12:51:30.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:30.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:52:41.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000289/group_15/index_8881_seed_10698_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 12:53:04.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.694 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 13:04:33.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_15/index_20_seed_22527_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 13:05:54.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:54.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:07.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000290/group_15/index_20_seed_22528_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:07:33.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 13:18:22.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_15/index_9921_seed_1014_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 13:19:44.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:44.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:20:53.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000291/group_15/index_9921_seed_1015_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:21:14.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 13:32:07.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_15/index_7965_seed_13044_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 13:33:31.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:31.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:38.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000292/group_15/index_7965_seed_13045_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:35:00.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 13:45:49.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:49.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_15/index_1302_seed_12310_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 13:47:09.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:47:09.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:22.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000293/group_15/index_1302_seed_12311_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 13:48:42.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 13:59:30.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:30.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_15/index_602_seed_11601_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 14:00:54.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:54.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:00.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000294/group_15/index_602_seed_11602_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:02:21.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 14:13:08.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:14.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_15/index_8376_seed_10867_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 14:14:29.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:29.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:15:42.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000295/group_15/index_8376_seed_10868_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:16:09.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 14:27:02.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_15/index_9794_seed_11637_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 14:28:22.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:22.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:31.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000296/group_15/index_9794_seed_11638_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:29:51.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 14:40:44.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_15/index_111_seed_15098_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 14:42:03.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:42:03.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:15.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000297/group_15/index_111_seed_15099_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:43:36.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 14:54:25.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:31.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_15/index_4525_seed_5080_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 14:55:45.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:56:55.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000298/group_15/index_4525_seed_5081_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 14:57:15.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 15:08:07.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_15/index_6956_seed_17019_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 15:09:29.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:29.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:10:39.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000299/group_15/index_6956_seed_17020_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:11:02.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.024 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 15:23:36.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_15/index_6999_seed_8553_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 15:24:54.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:54.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:08.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000300/group_15/index_6999_seed_8554_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:26:31.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 15:37:21.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:27.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_15/index_6415_seed_1410_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 15:38:39.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:39.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:39:52.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000301/group_15/index_6415_seed_1411_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:40:14.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 15:51:08.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:14.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_15/index_7241_seed_6550_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 15:52:28.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:28.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:53:40.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000302/group_15/index_7241_seed_6551_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 15:54:05.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 16:04:54.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_15/index_863_seed_18534_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 16:06:19.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:06:19.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:24.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000303/group_15/index_863_seed_18535_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:07:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 16:18:41.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:41.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_15/index_5980_seed_25555_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 16:20:01.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:20:01.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:12.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000304/group_15/index_5980_seed_25556_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:21:34.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 16:32:23.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_15/index_7158_seed_18054_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 16:33:45.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:45.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:34:53.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000305/group_15/index_7158_seed_18055_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 16:35:13.431 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:35:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 16:46:07.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:13.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_15/index_3195_seed_30071_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 16:47:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:39.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000306/group_15/index_3195_seed_30072_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 16:49:00.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 16:59:50.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_15/index_960_seed_994_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 17:01:14.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:01:14.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:21.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000307/group_15/index_960_seed_995_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:02:44.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 17:13:34.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:40.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_15/index_4107_seed_1516_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 17:14:52.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:52.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:08.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000308/group_15/index_4107_seed_1517_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:16:31.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.293 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 17:27:21.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_15/index_8942_seed_8955_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 17:28:42.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:42.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:29:53.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000309/group_15/index_8942_seed_8956_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:30:18.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.887 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 17:41:44.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_15/index_6373_seed_26584_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 17:43:04.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:43:04.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:21.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000310/group_15/index_6373_seed_26585_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:44:43.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 17:55:31.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:37.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_15/index_2678_seed_2699_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 17:56:54.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:54.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:02.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000311/group_15/index_2678_seed_2700_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 17:58:25.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 18:09:15.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:15.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_15/index_9636_seed_31098_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 18:10:34.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:34.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:11:49.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000312/group_15/index_9636_seed_31099_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:12:10.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.372 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 18:23:03.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:09.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_15/index_4912_seed_5846_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 18:24:23.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:23.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:33.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000313/group_15/index_4912_seed_5847_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:25:54.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.092 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 18:36:44.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_15/index_2989_seed_23432_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 18:38:07.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:38:07.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:16.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000314/group_15/index_2989_seed_23433_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:39:39.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.127 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 18:50:26.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:32.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_15/index_6928_seed_28279_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 18:51:54.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:54.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:52:59.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000315/group_15/index_6928_seed_28280_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 18:53:22.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.933 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 19:04:13.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_15/index_8046_seed_531_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 19:05:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:06:45.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000316/group_15/index_8046_seed_532_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:07:08.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 19:18:02.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_15/index_1179_seed_26451_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 19:19:23.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:23.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:30.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000317/group_15/index_1179_seed_26452_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:20:53.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 19:31:44.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:50.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_15/index_6479_seed_30931_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 19:33:04.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:33:04.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:16.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000318/group_15/index_6479_seed_30932_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:34:37.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 19:45:24.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_15/index_2330_seed_2888_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 19:46:51.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:51.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:47:56.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000319/group_15/index_2330_seed_2889_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 19:48:18.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.704 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 19:59:44.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:50.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_15/index_9701_seed_12403_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 20:01:05.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:01:05.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:15.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000320/group_15/index_9701_seed_12404_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 20:02:34.357 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:02:35.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 20:13:25.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:25.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:30.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_15/index_58_seed_29251_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 20:14:43.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:43.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:15:55.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000321/group_15/index_58_seed_29252_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:16:16.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 20:27:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_15/index_2998_seed_11041_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 20:28:29.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:29.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:38.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000322/group_15/index_2998_seed_11042_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:29:59.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 20:40:49.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_15/index_3960_seed_9076_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 20:42:10.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:42:10.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:22.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000323/group_15/index_3960_seed_9077_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:43:46.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 20:54:36.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:42.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_15/index_5106_seed_8275_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 20:55:59.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:59.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:09.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000324/group_15/index_5106_seed_8276_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 20:57:34.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 21:09:41.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_15/index_8946_seed_3290_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 21:11:01.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:11:01.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:14.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000325/group_15/index_8946_seed_3291_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:12:35.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 21:23:24.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_15/index_6411_seed_7729_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 21:24:47.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:47.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:25:53.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000326/group_15/index_6411_seed_7730_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:26:15.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 21:37:07.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:13.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_15/index_302_seed_6564_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 21:38:29.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:29.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:36.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000327/group_15/index_302_seed_6565_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:39:58.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 21:50:48.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_15/index_3141_seed_2990_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 21:52:11.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:52:11.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:21.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000328/group_15/index_3141_seed_2991_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 21:53:42.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 21:53:43.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 22:04:31.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_15/index_7228_seed_20251_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 22:05:54.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:54.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:04.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000329/group_15/index_7228_seed_20252_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:07:26.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.777 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 22:18:53.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:59.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_15/index_7281_seed_24160_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 22:20:11.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:20:11.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:24.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000330/group_15/index_7281_seed_24161_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:21:47.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 22:32:40.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:46.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_15/index_4964_seed_31976_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 22:34:02.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:34:02.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:12.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000331/group_15/index_4964_seed_31977_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:35:35.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 22:46:24.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_15/index_8453_seed_1573_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 22:47:42.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:42.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:48:56.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000332/group_15/index_8453_seed_1574_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 22:49:18.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.311 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.281 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 23:00:07.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_15/index_264_seed_2209_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 23:01:28.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:28.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:38.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000333/group_15/index_264_seed_2210_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:02:59.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 23:13:49.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:55.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_15/index_8168_seed_629_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 23:15:09.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:15:09.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:20.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000334/group_15/index_8168_seed_630_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:16:44.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 23:27:32.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:38.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_15/index_3971_seed_31304_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 23:28:53.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:53.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:02.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000335/group_15/index_3971_seed_31305_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:30:23.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 23:41:14.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:20.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_15/index_9967_seed_15513_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 23:42:40.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:40.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:43:44.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000336/group_15/index_9967_seed_15514_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:44:06.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-28 23:55:00.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:06.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_15/index_7428_seed_21667_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-28 23:56:21.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:21.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:29.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000337/group_15/index_7428_seed_21668_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-28 23:57:50.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-29 00:08:39.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:45.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_15/index_3544_seed_10596_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-29 00:09:59.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:59.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:08.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000338/group_15/index_3544_seed_10597_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:11:29.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-29 00:22:19.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:25.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_15/index_6058_seed_27050_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-29 00:23:39.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:39.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:24:52.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000339/group_15/index_6058_seed_27051_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:25:13.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.219 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.403 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-29 00:36:43.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_15/index_1598_seed_1618_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-29 00:38:03.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:38:03.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:15.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000340/group_15/index_1598_seed_1619_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:39:39.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-29 00:50:27.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:33.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_15/index_1186_seed_31175_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-29 00:51:49.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:49.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:52:56.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000341/group_15/index_1186_seed_31176_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 00:53:19.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 62, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 62, dp_rank 31: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=15, rank_in_group=1 +2026-06-29 01:04:09.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:15.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_15/index_9112_seed_9334_dp_31_rank_62_sample_2_slot_0.mp4 +2026-06-29 01:05:32.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 62 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:32.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 62 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:06:41.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1571 - Rank 62: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000342/group_15/index_9112_seed_9335_dp_31_rank_62_sample_3_slot_1.mp4 +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 62: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +2026-06-29 01:07:04.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 62: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 62: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 62, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_7.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_7.log new file mode 100644 index 0000000000000000000000000000000000000000..2f1dd795449262082284d6b29cc51b1d5c995d97 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_7.log @@ -0,0 +1,5147 @@ +2026-05-29 20:45:53.338 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:45:53.339 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 20:51:46.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:11.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:11.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.829 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 7, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 21:06:57.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:01.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 21:21:17.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:15.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 21:35:35.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:43.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:43.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.112 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 21:49:56.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:04.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.288 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 22:04:18.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.373 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 22:18:35.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:41.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 22:32:58.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:58.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:00.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.542 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 22:47:20.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:27.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:28.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.043 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 23:01:44.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:51.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:51.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:48.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:53.036 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 23:16:57.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:57.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 23:31:14.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 23:45:29.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:35.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:35.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-29 23:59:50.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:50.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:57.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:57.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.222 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.956 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 00:14:13.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 00:28:36.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:44.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:44.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.884 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 00:42:58.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:06.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 00:57:15.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:21.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:21.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 01:11:38.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.923 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 01:26:05.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.674 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 01:41:19.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:27.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:27.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:18.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 01:55:38.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:44.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.589 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:39.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.048 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 02:09:58.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:02.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 02:24:23.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:31.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 02:38:56.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:02.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:02.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:57.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 02:54:31.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:37.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:37.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.214 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 03:08:50.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:57.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:57.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:53.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 03:23:09.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:16.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:16.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.442 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.668 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.427 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 03:37:26.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.277 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 03:51:59.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:05.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:05.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.476 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 04:07:19.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 04:21:48.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:58.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:58.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 04:36:12.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 04:50:41.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.265 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 05:05:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:13.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:12.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 05:19:25.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 05:33:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:44.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:44.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:37.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 05:47:59.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:58.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 06:02:21.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.720 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 06:16:55.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.811 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 06:19:59.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:11.925 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 06:32:17.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:16.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 06:46:36.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:43.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:38.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 07:00:59.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:07.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:00.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.543 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 07:15:21.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:21.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:41.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 07:30:08.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 07:44:30.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:37.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:37.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 07:59:00.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.901 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:08.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.809 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 08:13:33.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:40.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:40.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 08:28:15.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:23.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:23.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.688 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:23.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 08:42:53.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:01.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:01.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:24.977 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 08:59:29.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:39.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.037 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:33.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 09:14:04.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:11.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:16.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.550 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 09:28:42.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:51.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 09:43:17.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.025 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:24.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 09:57:46.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:53.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 10:12:07.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.855 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:06.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.892 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 10:26:24.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:31.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:31.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:44.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 10:40:46.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:53.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:53.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 10:55:08.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:16.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:16.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 11:09:40.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:48.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:48.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.201 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:03.207 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 11:25:08.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:17.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:17.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.076 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:21.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 11:39:42.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:50.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 11:54:06.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 12:08:30.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:37.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:29.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 12:22:53.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:01.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:01.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 12:37:12.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.052 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:11.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 12:51:30.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.993 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 13:05:59.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:05.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:05.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:04.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.560 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 13:20:24.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:31.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:31.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:30.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 13:34:48.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.661 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 13:50:02.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:11.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.875 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 14:04:33.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:34.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 14:19:43.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:54.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:54.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 14:34:38.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 14:49:20.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:27.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.226 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 15:05:01.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:09.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:03.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 15:19:35.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 15:34:11.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:20.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:20.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.287 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 15:48:48.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:50.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 16:03:17.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:24.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:24.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.317 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 16:18:48.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:01.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.345 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 16:33:33.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 16:48:15.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 17:02:55.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.377 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 17:17:35.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:43.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:43.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.722 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 17:32:10.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 17:46:54.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:03.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:03.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 18:01:35.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.549 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 18:16:17.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:17.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 18:31:18.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.308 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:18.128 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 18:47:24.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:39.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 19:02:13.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:34.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 19:17:12.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:26.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:53.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.355 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 19:31:55.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 19:47:03.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:11.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:11.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.584 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.982 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 20:01:46.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.383 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 20:16:29.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 20:31:19.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:19.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 20:46:03.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:11.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:11.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 21:00:47.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:47.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.501 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 21:17:31.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:40.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:40.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:37.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 21:32:08.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:48:41.476 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:48:41.477 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 21:55:22.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:48.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:48.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:08.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.886 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 7, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 22:10:52.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:12:00.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:12:00.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 22:25:33.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:41.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:41.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 22:40:03.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:11.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:11.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:10.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 22:54:38.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:38.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:46.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:50.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 23:09:20.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:27.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:27.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.171 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.968 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 23:24:05.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:14.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 23:38:46.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:54.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-30 23:53:10.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:17.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:17.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:15.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 00:07:41.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:52.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:52.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:54.143 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 00:22:59.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.920 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 00:37:30.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:38.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:38.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 00:52:07.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:13.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:13.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:22.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 01:06:44.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:52.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:52.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 01:21:16.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:16.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.750 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 01:35:36.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:43.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:43.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 01:50:03.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:10.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:06.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 02:04:34.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:42.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.764 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:39.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 7, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-05-31 02:19:01.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:07.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:07.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:02.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:00:42.627 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:00:42.630 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 18:04:09.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:30.125 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:30.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 7, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 18:18:41.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:35.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 18:32:26.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.662 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 18:46:08.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:12.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:12.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:03.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 19:00:02.864 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:02.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 19:13:49.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 19:27:32.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:24.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 19:41:16.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:16.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:21.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:21.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.972 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 19:55:08.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:13.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:13.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.111 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 20:09:02.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.727 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:29.954 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 20:23:32.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.761 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.625 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 20:37:18.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:18.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:23.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:23.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 20:51:03.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:57.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.058 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 21:04:50.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:56.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.685 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 21:18:43.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:48.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 21:32:31.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.628 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 21:46:16.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:04.975 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 22:00:05.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:11.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 22:13:50.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:56.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:56.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 22:27:39.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:44.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.089 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:58.389 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 22:41:59.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:42:00.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 22:55:45.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:50.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.618 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 23:09:30.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:35.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 23:23:15.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:20.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:11.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 23:37:07.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:12.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:01.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-25 23:51:52.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 00:05:36.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 00:19:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:26.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 00:33:07.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 00:47:03.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:00.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.434 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.529 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:29.170 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 01:01:30.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 01:15:17.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:23.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:12.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 01:29:08.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:13.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:13.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 01:42:58.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 01:56:44.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.165 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 02:10:30.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:30.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 02:24:15.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 02:38:08.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:13.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.094 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 02:40:59.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 02:51:53.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:58.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:58.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.940 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 03:05:40.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:46.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:04.827 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 03:20:06.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 03:33:52.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:57.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:57.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.186 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 03:47:40.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:45.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 04:01:21.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.429 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 04:15:09.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.144 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 04:29:02.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 04:42:46.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.440 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 04:56:27.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:32.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:32.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.656 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 05:10:10.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:03.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 05:24:02.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.569 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.206 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 05:39:29.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:35.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.09s (1.87min) +2026-06-26 05:44:13.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 05:53:10.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:16.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:16.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 06:06:52.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 06:20:39.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:45.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 06:34:21.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:26.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:26.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.499 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.930 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.616 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 06:48:09.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:15.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:15.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 07:02:03.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.348 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:55.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 07:15:48.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.420 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 07:29:32.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:37.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:30.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 07:43:21.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:44.055 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 07:57:45.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:52.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 08:11:33.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 08:25:18.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:08.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 08:39:03.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:57.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 08:52:49.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:55.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.234 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 09:06:39.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 09:20:22.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:27.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:27.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 09:34:09.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:15.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:15.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.159 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 09:47:56.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:02.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.361 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:51.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 10:01:45.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:35.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.820 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.937 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 10:16:01.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:01.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:59.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.274 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.680 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.638 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 10:29:52.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:44.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 10:43:37.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.281 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:29.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 10:57:21.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 11:11:08.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:14.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:14.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 11:26:08.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:13.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.142 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 11:39:55.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 11:53:37.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:41.761 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:41.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:31.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.494 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 12:07:21.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.286 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:12.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 12:21:06.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:06.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:28.814 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 12:35:30.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:37.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:37.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 12:49:16.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:21.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.744 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.064 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 13:03:03.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:08.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:53.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 13:16:44.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:49.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 13:30:31.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 13:44:15.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.284 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 13:58:07.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.577 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 14:11:49.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:39.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 14:25:28.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:33.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:25.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 14:39:15.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:20.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:20.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:13.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.442 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 14:53:45.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.366 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 15:07:27.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 15:21:11.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:16.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.111 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 15:34:56.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.264 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.564 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 15:48:45.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:50.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.312 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 16:02:26.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:31.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:31.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.921 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:17.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:05.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 16:16:06.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:12.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:04.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 16:29:54.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:30:59.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:30:59.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.930 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 16:43:38.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.655 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 16:57:20.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:20.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:25.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:25.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.142 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 17:12:53.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.055 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:48.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 17:26:39.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:44.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 17:40:20.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 17:54:06.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 18:07:49.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.980 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 18:21:38.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:37.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.846 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.994 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 18:35:27.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:27.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:32.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:32.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.637 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.294 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 18:49:15.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:20.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:05.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.890 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 19:03:01.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:02.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:07.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 19:16:46.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:51.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:30.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.469 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 19:31:08.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:07.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 19:45:03.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 19:58:45.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:46.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:51.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:37.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.831 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 20:12:28.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.128 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:22.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 20:26:14.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 20:40:06.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:11.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:11.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.483 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 20:53:52.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:58.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.176 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:54.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.906 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 21:07:44.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.549 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.818 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.479 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 21:21:31.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.688 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 21:35:16.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:22.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:11.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.563 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 21:49:43.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:44.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.490 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 22:03:26.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:31.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:31.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:17.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 22:17:07.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 22:30:49.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 22:44:32.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.769 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 22:59:16.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.729 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 23:13:02.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:07.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:07.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:53.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 23:26:44.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 23:40:27.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:26.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-26 23:54:16.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:21.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:38.128 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 00:08:40.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 00:22:20.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.490 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.965 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.686 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 00:36:06.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:59.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 00:49:49.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.944 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.585 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 01:03:35.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:40.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.584 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.033 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 01:17:19.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:19.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:25.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 01:31:06.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:12.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:58.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.908 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 01:44:47.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:52.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:52.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 01:58:27.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 02:12:14.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.540 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 02:26:39.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:46.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.168 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 02:40:23.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:28.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.774 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 02:54:14.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:09.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.445 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 03:08:01.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:01.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:07.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.205 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:50.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 03:21:41.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:47.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:47.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 03:35:26.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 03:49:14.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.393 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 04:03:03.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:08.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:08.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 04:16:48.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:53.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.073 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 04:30:30.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:30.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:36.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:36.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.552 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.588 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 04:45:58.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:04.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:04.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 04:59:46.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:52.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:40.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 05:13:27.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:32.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:32.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:10.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 05:27:07.942 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:08.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:13.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.418 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.549 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 05:40:50.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 05:54:35.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:39.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:39.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.464 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 06:08:19.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:14.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 06:22:08.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:13.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.189 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 06:35:50.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:55.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 06:49:30.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:30.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.916 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:12.055 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.867 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 07:03:49.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:56.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 07:17:29.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:34.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:22.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.632 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 07:31:16.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:12.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.067 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 07:45:07.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 07:58:52.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:57.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.143 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 08:12:37.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:43.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 08:26:20.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:11.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.138 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 08:40:02.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:08.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.092 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 08:53:47.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:47.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:53.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.418 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 09:07:41.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.888 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:46.892 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.443 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.640 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:00.934 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 09:22:02.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 09:35:43.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:48.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 09:49:26.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:31.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:18.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.244 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 10:03:08.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:13.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:13.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.862 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.251 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 10:16:53.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:58.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.587 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.600 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 10:31:49.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:49.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:54.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:54.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 10:45:31.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.262 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.758 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 10:59:16.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:21.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:21.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:12.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.388 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 11:13:07.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:12.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 11:26:47.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.798 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:42.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.509 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.478 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 11:41:06.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:12.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:59.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.347 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.130 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 11:54:48.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:53.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 12:08:34.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:39.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 12:22:19.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.431 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.579 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.502 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 12:36:01.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 12:49:44.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.554 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 13:03:21.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 13:17:07.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.462 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:13.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 13:30:47.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:47.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.416 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 13:44:38.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:43.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.191 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 13:58:57.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:02.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.266 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.483 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 14:12:43.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.157 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 14:26:20.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:25.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:12.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:05.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 14:40:05.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:05.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:11.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:58.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 14:53:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.811 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 15:07:26.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:27.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:32.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.903 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 15:21:06.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:11.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:11.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:58.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 15:34:47.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:53.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 15:48:32.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:37.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 16:02:14.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:05.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.690 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 16:17:43.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:44.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:49.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:49.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 16:31:26.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 16:45:07.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:59.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 16:58:47.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:52.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:42.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 17:12:31.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 17:26:15.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:20.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:09.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.061 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 17:40:00.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:00.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:05.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:05.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 17:53:45.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:38.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 18:07:25.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:24.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 18:21:14.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:19.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:19.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:08.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.808 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 18:35:38.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:43.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:43.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.570 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.131 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 18:49:21.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 19:03:06.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:11.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:11.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.631 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 19:16:46.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:51.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 19:30:30.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:25.147 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.317 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.235 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 19:44:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:18.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:18.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.992 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:10.446 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 19:58:02.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:53.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 20:11:43.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:44.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:49.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:39.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 20:25:26.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:19.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 20:39:06.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:06.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:12.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.410 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.438 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.228 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 20:53:29.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:34.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.306 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:21.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 21:07:10.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:16.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:16.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:06.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 21:21:01.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:51.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.974 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 21:34:42.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 21:48:24.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.458 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 22:03:21.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:16.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 22:17:06.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:07.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.152 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:57.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.444 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 22:30:47.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:52.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 22:44:30.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.466 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.485 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 22:58:07.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:12.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.656 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 23:12:29.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:36.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:36.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 23:26:15.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:20.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 23:40:02.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.290 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-27 23:53:51.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:55.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:55.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 00:07:32.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.225 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.020 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 00:21:21.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:21.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:27.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 00:35:07.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 00:49:02.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:02.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:07.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:07.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:57.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.011 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 01:02:48.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 01:16:33.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.629 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 01:30:58.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.679 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.880 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 01:44:44.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.772 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.691 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 01:58:27.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.693 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 02:12:08.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:14.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:14.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:04.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.567 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 02:25:53.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.924 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 02:39:38.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:31.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 02:53:20.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:20.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:25.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:25.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.403 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 02:58:07.354 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 03:07:08.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:03.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 03:20:53.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:49.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.593 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 03:34:40.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:45.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.480 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.813 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:12.296 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 03:50:13.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:13.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.802 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 04:04:03.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.563 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 04:17:48.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:54.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 04:31:40.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 04:45:20.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:10.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 04:59:03.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 05:12:44.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:44.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:49.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:49.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:35.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.172 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 05:26:26.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:22.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 05:40:15.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.828 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.118 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 05:54:06.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:12.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.959 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 06:08:27.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:21.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.630 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 06:22:11.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:16.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:16.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:07.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.715 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 06:36:02.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:07.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:07.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.836 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 06:49:45.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.729 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.807 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:26.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 07:03:27.755 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.148 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 07:17:12.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:17.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:04.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.665 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.960 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 07:30:53.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:48.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 07:44:40.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:32.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 07:58:21.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:26.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:26.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 08:12:08.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:14.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:14.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.472 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.745 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 08:26:31.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.234 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 08:40:22.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:28.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:14.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 08:54:07.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:13.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 08:57:00.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.059 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 09:07:50.860 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:50.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:55.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:55.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.674 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 09:21:32.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:37.669 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:37.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 09:36:35.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.710 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.732 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 09:50:22.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:25.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 10:04:16.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 10:18:09.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:15.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:15.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:02.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 10:31:54.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.259 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:54.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.772 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 10:46:20.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 11:00:08.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.508 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.368 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 11:13:51.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 11:27:32.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.898 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 11:41:15.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.475 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 11:55:03.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.899 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 12:08:48.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:53.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:53.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:48.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.792 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 12:22:40.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.817 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 12:36:24.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 12:50:08.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:14.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.765 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.889 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 13:04:33.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.018 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:34.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 13:18:22.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:28.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:14.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.320 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 13:32:07.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.611 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 13:45:49.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:54.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:54.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.544 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 13:59:31.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:36.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:21.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.180 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.561 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 14:13:08.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 14:27:02.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.158 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 14:40:44.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:48.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:48.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 14:54:25.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 15:08:07.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:12.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:02.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.817 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.332 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 15:23:36.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:31.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 15:37:21.290 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 15:51:08.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 16:04:54.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:50.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:39.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 16:18:40.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:45.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:45.989 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 16:32:23.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:28.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.523 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 16:46:07.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.322 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 16:59:50.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.284 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 17:13:34.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 17:27:21.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:26.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.659 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 17:41:44.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:49.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:49.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 17:55:31.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:32.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.794 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 18:09:15.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:16.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.171 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:21.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 18:23:03.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 18:36:44.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:45.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:50.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:39.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 18:50:26.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:22.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.076 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 19:04:13.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:13.925 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:19.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.109 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:08.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 19:18:01.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:07.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.248 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.589 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 19:31:44.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.662 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.366 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 19:45:24.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:24.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:29.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:29.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.478 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 19:59:44.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 20:13:24.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:24.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:29.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:29.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:16.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 20:27:07.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:12.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 20:40:49.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:54.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.202 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 20:54:36.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.482 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 21:09:41.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:46.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:46.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 21:23:24.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.107 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:29.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.845 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.398 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 21:37:07.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:47.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 21:50:48.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:54.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 22:04:31.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:37.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.734 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 22:18:53.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.144 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 22:32:40.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:35.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.151 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.340 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 22:46:24.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:29.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.443 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 23:00:07.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.289 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.268 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:59.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 23:13:49.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:50.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:54.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 23:27:32.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:23.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 23:41:14.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-28 23:55:00.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.648 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-29 00:08:39.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:29.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-29 00:22:19.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.350 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-29 00:36:43.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:48.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.559 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.468 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-29 00:50:27.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.756 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 7, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 7, dp_rank 3: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=1, rank_in_group=1 +2026-06-29 01:04:09.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:15.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 7 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:15.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 7 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 7: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:03.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 7: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 7: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 7, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_9.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_9.log new file mode 100644 index 0000000000000000000000000000000000000000..5f965ba9e5f1440df26186f7d1027c0082bb1931 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/logs/training_rank_9.log @@ -0,0 +1,5147 @@ +2026-05-29 20:47:42.434 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:47:42.434 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 20:51:46.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.348 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:53:11.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:11.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 20:55:17.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 20:57:22.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.811 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 20:59:26.610 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 9, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-29 21:06:56.429 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 1, train_steps: 40 +2026-05-29 21:06:57.334 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 21:06:57.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:06:57.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:08:03.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:08:03.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:10:00.943 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 21:10:01.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:12:04.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-29 21:12:04.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:14:07.125 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:21:16.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 2, train_steps: 80 +2026-05-29 21:21:17.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 21:21:17.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:21:17.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:22:24.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:22:24.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:24:14.955 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 21:24:15.791 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:26:19.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.63s (2.04min) +2026-05-29 21:26:19.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:28:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:35:34.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 3, train_steps: 120 +2026-05-29 21:35:35.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 21:35:35.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:35:35.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:36:42.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:36:42.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:38:35.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 21:38:36.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:40:37.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-29 21:40:38.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:42:42.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 21:49:55.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 4, train_steps: 160 +2026-05-29 21:49:56.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 21:49:56.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:49:56.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:51:04.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 21:51:04.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 21:53:00.467 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 21:53:01.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 21:55:05.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.64s (2.06min) +2026-05-29 21:55:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 21:57:06.289 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:04:17.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 5, train_steps: 200 +2026-05-29 22:04:18.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 22:04:18.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:04:18.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:05:25.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:05:25.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:07:14.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 22:07:15.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:09:17.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.07s (2.03min) +2026-05-29 22:09:18.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:11:21.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:18:34.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 6, train_steps: 240 +2026-05-29 22:18:35.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 22:18:35.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:18:35.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:19:41.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:19:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:21:37.289 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 22:21:38.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:23:43.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.55s (2.08min) +2026-05-29 22:23:44.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:25:46.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:32:57.789 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 7, train_steps: 280 +2026-05-29 22:32:58.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 22:32:59.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:32:59.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:34:05.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:34:05.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:35:59.983 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 22:36:00.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:38:04.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.09s (2.05min) +2026-05-29 22:38:05.542 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:40:08.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 22:47:19.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 8, train_steps: 320 +2026-05-29 22:47:20.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 22:47:20.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:47:20.584 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:48:27.952 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 22:48:27.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 22:50:19.326 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 22:50:20.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 22:52:27.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.20s (2.10min) +2026-05-29 22:52:28.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 22:54:29.730 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:01:43.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 9, train_steps: 360 +2026-05-29 23:01:43.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 23:01:44.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:01:44.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:02:51.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:02:51.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:04:48.130 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 23:04:48.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:06:53.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-29 23:06:54.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:08:54.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:16:04.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 10, train_steps: 400 +2026-05-29 23:16:52.952 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-29 23:16:57.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 23:16:57.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:16:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:18:05.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:18:05.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:19:57.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 23:19:57.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:22:00.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.80s (2.03min) +2026-05-29 23:22:01.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:24:03.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:31:12.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 11, train_steps: 440 +2026-05-29 23:31:14.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 23:31:14.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:31:14.327 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:32:20.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:32:20.689 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:34:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 23:34:15.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:36:17.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.95s (2.02min) +2026-05-29 23:36:17.881 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:38:17.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:45:28.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 12, train_steps: 480 +2026-05-29 23:45:29.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 23:45:29.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:45:29.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:46:36.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:46:36.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-29 23:48:30.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-29 23:48:31.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-29 23:50:34.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.78s (2.05min) +2026-05-29 23:50:35.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 23:52:36.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-29 23:59:49.755 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 13, train_steps: 520 +2026-05-29 23:59:50.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-29 23:59:50.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 23:59:50.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:00:56.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:00:57.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:02:54.223 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 00:02:55.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:04:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.57s (2.06min) +2026-05-30 00:05:00.253 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:07:01.955 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:14:12.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 14, train_steps: 560 +2026-05-30 00:14:13.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 00:14:13.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:14:13.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:15:19.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:15:19.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:17:17.156 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 00:17:18.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:19:19.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.17s (2.02min) +2026-05-30 00:19:20.588 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:21:21.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:28:34.700 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 15, train_steps: 600 +2026-05-30 00:28:35.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 00:28:36.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:28:36.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:29:45.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:29:45.158 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:31:38.350 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 00:31:39.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:33:44.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.70s (2.08min) +2026-05-30 00:33:45.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:35:44.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:42:56.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 16, train_steps: 640 +2026-05-30 00:42:57.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 00:42:58.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:42:58.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:44:06.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:44:06.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:46:01.224 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 00:46:02.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 00:48:03.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.06s (2.02min) +2026-05-30 00:48:04.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 00:50:04.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 00:57:14.546 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 17, train_steps: 680 +2026-05-30 00:57:15.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 00:57:15.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:57:15.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 00:58:21.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 00:58:21.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:00:15.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 01:00:16.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:02:20.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.04s (2.05min) +2026-05-30 01:02:21.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:04:24.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:11:37.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 18, train_steps: 720 +2026-05-30 01:11:38.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 01:11:38.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:11:38.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:12:44.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:12:44.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:14:42.924 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 01:14:44.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:16:46.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.83s (2.03min) +2026-05-30 01:16:47.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:18:52.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:26:04.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 19, train_steps: 760 +2026-05-30 01:26:05.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 01:26:05.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:26:05.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:27:13.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:27:13.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:29:03.428 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 01:29:04.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:31:09.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.72s (2.06min) +2026-05-30 01:31:10.012 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:33:09.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:40:21.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 20, train_steps: 800 +2026-05-30 01:41:09.312 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-05-30 01:41:19.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 01:41:19.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:41:19.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:42:27.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:42:27.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:44:17.733 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 01:44:18.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 01:46:20.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.30s (2.02min) +2026-05-30 01:46:21.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 01:48:23.514 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 01:55:37.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 21, train_steps: 840 +2026-05-30 01:55:37.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 01:55:38.082 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:55:38.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:56:44.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 01:56:44.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 01:58:38.590 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 01:58:39.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:00:44.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.79s (2.06min) +2026-05-30 02:00:44.895 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:02:44.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:09:57.063 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 22, train_steps: 880 +2026-05-30 02:09:57.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 02:09:58.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:09:58.231 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:11:04.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:11:04.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:13:01.915 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 02:13:02.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:15:07.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.90s (2.06min) +2026-05-30 02:15:08.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:17:10.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:24:22.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 23, train_steps: 920 +2026-05-30 02:24:23.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 02:24:23.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:24:23.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:25:31.617 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:25:31.621 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:27:32.647 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 02:27:33.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:29:38.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.73s (2.08min) +2026-05-30 02:29:39.775 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:31:42.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:38:55.457 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 24, train_steps: 960 +2026-05-30 02:38:56.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 02:38:56.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:38:56.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:40:02.790 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:40:02.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:41:57.062 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 02:41:58.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:44:00.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 02:44:01.678 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 02:46:04.009 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 02:53:15.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 25, train_steps: 1000 +2026-05-30 02:54:31.440 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 02:54:31.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:54:31.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:55:37.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 02:55:37.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 02:57:30.215 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 02:57:31.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 02:59:33.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.82s (2.03min) +2026-05-30 02:59:34.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:01:36.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:08:49.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 26, train_steps: 1040 +2026-05-30 03:08:50.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 03:08:50.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:08:50.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:09:56.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:09:56.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:11:52.553 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 03:11:53.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:13:56.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.67s (2.04min) +2026-05-30 03:13:57.876 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:15:58.685 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:23:08.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 27, train_steps: 1080 +2026-05-30 03:23:09.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 03:23:09.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:23:09.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:24:16.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:24:16.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:26:07.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 03:26:08.305 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:28:11.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.35s (2.04min) +2026-05-30 03:28:12.240 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:30:15.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:37:25.430 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 28, train_steps: 1120 +2026-05-30 03:37:26.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 03:37:26.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:37:26.831 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:38:33.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:38:33.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:40:38.278 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 03:40:39.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:42:44.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.84s (2.08min) +2026-05-30 03:42:45.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:44:44.976 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 03:51:58.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 29, train_steps: 1160 +2026-05-30 03:51:59.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 03:51:59.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:51:59.518 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:53:06.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 03:53:06.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 03:55:03.449 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 03:55:04.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 03:57:08.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.78s (2.06min) +2026-05-30 03:57:09.946 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 03:59:12.312 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:06:24.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 30, train_steps: 1200 +2026-05-30 04:07:14.409 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-05-30 04:07:19.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 04:07:19.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:07:19.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:08:27.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:08:27.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:10:27.241 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 04:10:28.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:12:29.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.03s (2.02min) +2026-05-30 04:12:30.851 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:14:32.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:21:47.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 31, train_steps: 1240 +2026-05-30 04:21:48.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 04:21:48.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:21:48.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:22:57.947 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:22:57.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:24:53.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 04:24:54.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:26:57.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.24s (2.04min) +2026-05-30 04:26:58.170 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:28:58.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:36:11.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 32, train_steps: 1280 +2026-05-30 04:36:12.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 04:36:12.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:36:12.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:37:19.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:37:19.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:39:15.343 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 04:39:16.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:41:21.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.01s (2.08min) +2026-05-30 04:41:22.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:43:25.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 04:50:40.673 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 33, train_steps: 1320 +2026-05-30 04:50:41.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 04:50:41.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:50:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:51:47.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 04:51:47.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 04:53:43.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 04:53:44.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 04:55:48.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.84s (2.06min) +2026-05-30 04:55:49.508 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 04:57:55.176 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:05:05.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 34, train_steps: 1360 +2026-05-30 05:05:06.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 05:05:07.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:05:07.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:06:13.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:06:13.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:08:12.020 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 05:08:12.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:10:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.00s (2.02min) +2026-05-30 05:10:15.049 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:12:13.793 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:19:24.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 35, train_steps: 1400 +2026-05-30 05:19:25.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 05:19:25.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:19:25.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:20:31.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:20:31.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:22:19.803 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 05:22:20.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:24:23.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.57s (2.04min) +2026-05-30 05:24:24.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:26:23.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:33:37.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 36, train_steps: 1440 +2026-05-30 05:33:38.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 05:33:38.454 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:33:38.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:34:44.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:34:44.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:36:36.771 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 05:36:37.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:38:39.894 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.79s (2.03min) +2026-05-30 05:38:40.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:40:43.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 05:47:58.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 37, train_steps: 1480 +2026-05-30 05:47:59.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 05:47:59.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:47:59.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:49:06.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 05:49:06.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 05:50:57.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 05:50:58.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 05:53:02.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.36s (2.06min) +2026-05-30 05:53:03.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 05:55:05.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:02:20.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 38, train_steps: 1520 +2026-05-30 06:02:20.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 06:02:21.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:02:21.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:03:29.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:03:29.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:05:32.140 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 06:05:33.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:07:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.52s (2.01min) +2026-05-30 06:07:34.913 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:09:35.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:16:53.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 39, train_steps: 1560 +2026-05-30 06:16:54.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 06:16:55.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:16:55.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:18:02.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:18:02.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:19:58.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 06:19:59.803 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:22:04.534 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.12s (2.07min) +2026-05-30 06:22:05.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:24:06.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:31:22.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 40, train_steps: 1600 +2026-05-30 06:32:11.717 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-05-30 06:32:17.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 06:32:17.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:32:17.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:33:25.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:33:25.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:35:15.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 06:35:16.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:37:18.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.80s (2.01min) +2026-05-30 06:37:19.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:39:21.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 06:46:35.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 41, train_steps: 1640 +2026-05-30 06:46:36.858 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 06:46:36.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:46:37.017 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:47:43.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 06:47:43.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 06:49:37.064 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 06:49:38.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 06:51:38.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.25s (2.00min) +2026-05-30 06:51:39.531 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 06:53:42.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:00:58.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 42, train_steps: 1680 +2026-05-30 07:00:59.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 07:00:59.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:00:59.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:02:07.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:02:07.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:03:59.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 07:04:00.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:06:05.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.08s (2.07min) +2026-05-30 07:06:06.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:08:07.175 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:15:20.537 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 43, train_steps: 1720 +2026-05-30 07:15:21.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 07:15:22.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:15:22.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:16:28.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:16:28.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:18:40.788 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 07:18:41.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:20:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.39s (2.07min) +2026-05-30 07:20:47.879 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:22:52.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:30:07.378 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 44, train_steps: 1760 +2026-05-30 07:30:08.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 07:30:08.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:30:08.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:31:16.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:31:16.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:33:10.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 07:33:11.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:35:13.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.09s (2.02min) +2026-05-30 07:35:14.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:37:16.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:44:29.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 45, train_steps: 1800 +2026-05-30 07:44:30.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 07:44:30.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:44:30.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:45:37.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:45:37.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 07:47:29.561 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 07:47:30.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 07:49:33.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.33s (2.04min) +2026-05-30 07:49:34.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 07:51:38.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 07:58:58.945 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 46, train_steps: 1840 +2026-05-30 07:59:00.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 07:59:00.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 07:59:00.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:00:07.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:00:07.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:02:07.902 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 08:02:08.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:04:12.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.11s (2.05min) +2026-05-30 08:04:13.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:06:15.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:13:32.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 47, train_steps: 1880 +2026-05-30 08:13:33.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 08:13:33.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:13:34.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:14:40.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:14:40.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:16:43.123 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 08:16:44.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:18:48.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.00s (2.07min) +2026-05-30 08:18:49.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:20:57.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:28:14.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 48, train_steps: 1920 +2026-05-30 08:28:15.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 08:28:15.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:28:15.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:29:23.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:29:23.887 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:31:22.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 08:31:23.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:33:28.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.63s (2.06min) +2026-05-30 08:33:28.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:35:32.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:42:51.825 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 49, train_steps: 1960 +2026-05-30 08:42:52.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 08:42:53.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:42:53.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:44:02.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:44:02.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 08:46:00.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 08:46:01.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 08:48:06.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.97s (2.07min) +2026-05-30 08:48:07.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 08:50:07.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 08:57:23.135 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 50, train_steps: 2000 +2026-05-30 08:59:24.637 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-05-30 08:59:29.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 08:59:29.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 08:59:29.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:00:39.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:00:39.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:02:32.038 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 09:02:33.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:04:37.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.13s (2.07min) +2026-05-30 09:04:38.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:06:43.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:14:03.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 51, train_steps: 2040 +2026-05-30 09:14:04.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 09:14:04.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:14:05.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:15:11.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:15:11.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:17:15.997 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 09:17:17.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:19:18.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.26s (2.02min) +2026-05-30 09:19:19.551 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:21:24.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:28:41.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 52, train_steps: 2080 +2026-05-30 09:28:42.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 09:28:42.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:28:42.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:29:51.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:29:51.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:31:53.085 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 09:31:54.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:33:59.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.88s (2.08min) +2026-05-30 09:34:00.772 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:36:04.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:43:16.079 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 53, train_steps: 2120 +2026-05-30 09:43:17.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 09:43:17.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:43:17.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:44:25.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:44:25.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:46:24.026 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 09:46:25.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 09:48:26.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.76s (2.01min) +2026-05-30 09:48:27.066 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 09:50:29.218 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 09:57:44.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 54, train_steps: 2160 +2026-05-30 09:57:46.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 09:57:46.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:57:46.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 09:58:53.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 09:58:53.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:00:51.256 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 10:00:52.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:02:52.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.83s (2.00min) +2026-05-30 10:02:53.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:04:54.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:12:06.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 55, train_steps: 2200 +2026-05-30 10:12:07.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 10:12:07.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:12:07.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:13:15.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:13:15.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:15:05.856 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 10:15:06.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:17:08.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.52s (2.03min) +2026-05-30 10:17:09.856 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:19:10.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:26:22.901 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 56, train_steps: 2240 +2026-05-30 10:26:24.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 10:26:24.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:26:24.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:27:30.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:27:30.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:29:28.576 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 10:29:29.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:31:26.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 117.10s (1.95min) +2026-05-30 10:31:27.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:33:30.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:40:45.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 57, train_steps: 2280 +2026-05-30 10:40:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 10:40:46.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:40:46.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:41:54.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:41:54.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:43:54.066 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 10:43:55.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 10:45:54.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.08s (1.98min) +2026-05-30 10:45:55.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 10:47:56.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 10:55:07.422 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 58, train_steps: 2320 +2026-05-30 10:55:08.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 10:55:08.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:55:08.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:56:16.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 10:56:16.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 10:58:22.148 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 10:58:23.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:00:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.15s (1.97min) +2026-05-30 11:00:22.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:02:23.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:09:38.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 59, train_steps: 2360 +2026-05-30 11:09:40.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 11:09:40.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:09:40.581 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:10:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:10:48.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:12:42.202 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 11:12:43.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:14:46.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.70s (2.05min) +2026-05-30 11:14:47.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:16:52.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:24:07.777 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 60, train_steps: 2400 +2026-05-30 11:25:03.254 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-05-30 11:25:07.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 11:25:08.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:25:08.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:26:16.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:26:16.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:28:20.077 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 11:28:21.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:30:22.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.68s (2.01min) +2026-05-30 11:30:23.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:32:25.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:39:41.477 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 61, train_steps: 2440 +2026-05-30 11:39:42.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 11:39:42.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:39:42.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:40:50.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:40:50.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:42:41.218 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 11:42:42.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:44:44.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.63s (2.03min) +2026-05-30 11:44:45.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 11:46:47.707 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 11:54:04.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 62, train_steps: 2480 +2026-05-30 11:54:06.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 11:54:06.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:54:06.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:55:13.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 11:55:13.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 11:57:04.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 11:57:06.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 11:59:13.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.58s (2.11min) +2026-05-30 11:59:14.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:01:15.839 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:08:29.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 63, train_steps: 2520 +2026-05-30 12:08:30.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 12:08:30.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:08:30.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:09:37.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:09:37.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:11:28.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 12:11:30.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:13:31.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.11s (2.02min) +2026-05-30 12:13:32.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:15:34.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:22:52.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 64, train_steps: 2560 +2026-05-30 12:22:53.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 12:22:53.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:22:53.748 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:24:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:24:01.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:25:54.599 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 12:25:55.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:27:55.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 119.53s (1.99min) +2026-05-30 12:27:56.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:29:56.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:37:10.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 65, train_steps: 2600 +2026-05-30 12:37:11.958 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 12:37:12.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:37:12.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:38:19.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:38:19.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:40:10.053 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 12:40:10.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:42:12.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.49s (2.01min) +2026-05-30 12:42:13.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:44:13.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 12:51:29.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 66, train_steps: 2640 +2026-05-30 12:51:30.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 12:51:30.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:51:30.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:52:38.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 12:52:38.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 12:54:31.748 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 12:54:32.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 12:56:34.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.57s (2.03min) +2026-05-30 12:56:35.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 12:58:38.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:05:57.988 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 67, train_steps: 2680 +2026-05-30 13:05:59.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 13:05:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:05:59.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:07:06.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:07:06.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:09:03.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 13:09:04.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:11:05.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.64s (2.01min) +2026-05-30 13:11:06.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:13:08.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:20:23.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 68, train_steps: 2720 +2026-05-30 13:20:24.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 13:20:24.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:20:24.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:21:31.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:21:31.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:23:28.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 13:23:29.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:25:30.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 120.16s (2.00min) +2026-05-30 13:25:31.513 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:27:31.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:34:47.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 69, train_steps: 2760 +2026-05-30 13:34:48.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 13:34:48.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:34:48.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:35:55.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:35:55.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:37:47.780 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 13:37:48.636 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:39:52.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.97s (2.05min) +2026-05-30 13:39:53.261 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:41:53.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 13:49:06.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 70, train_steps: 2800 +2026-05-30 13:49:56.239 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-05-30 13:50:02.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 13:50:02.450 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:50:02.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:51:11.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 13:51:11.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 13:53:10.876 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 13:53:11.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 13:55:13.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.37s (2.02min) +2026-05-30 13:55:14.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 13:57:15.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:04:32.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 71, train_steps: 2840 +2026-05-30 14:04:33.700 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 14:04:33.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:04:33.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:05:43.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:05:43.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:07:58.730 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 14:07:59.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:10:10.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 130.46s (2.17min) +2026-05-30 14:10:11.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:12:20.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:19:42.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 72, train_steps: 2880 +2026-05-30 14:19:43.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 14:19:43.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:19:43.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:20:54.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:20:54.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:23:12.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 14:23:13.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:25:15.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 121.70s (2.03min) +2026-05-30 14:25:16.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:27:18.904 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:34:37.109 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 73, train_steps: 2920 +2026-05-30 14:34:38.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 14:34:38.544 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:34:38.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:35:45.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:35:45.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:37:52.938 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 14:37:54.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:40:01.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.71s (2.11min) +2026-05-30 14:40:02.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:42:05.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 14:49:18.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 74, train_steps: 2960 +2026-05-30 14:49:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 14:49:20.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:49:20.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:50:27.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 14:50:27.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 14:52:25.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 14:52:26.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 14:54:30.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.75s (2.05min) +2026-05-30 14:54:31.057 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 14:56:32.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:03:49.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 75, train_steps: 3000 +2026-05-30 15:05:01.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 15:05:01.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:05:01.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:06:09.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:06:09.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:08:02.892 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 15:08:04.085 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:10:11.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.72s (2.11min) +2026-05-30 15:10:12.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:12:17.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:19:33.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 76, train_steps: 3040 +2026-05-30 15:19:34.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 15:19:35.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:19:35.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:20:42.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:20:42.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:22:37.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 15:22:38.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:24:46.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.53s (2.13min) +2026-05-30 15:24:47.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:26:54.626 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:34:09.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 77, train_steps: 3080 +2026-05-30 15:34:10.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 15:34:11.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:34:11.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:35:20.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:35:20.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:37:18.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 15:37:19.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:39:23.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.31s (2.07min) +2026-05-30 15:39:24.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:41:30.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 15:48:47.299 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 78, train_steps: 3120 +2026-05-30 15:48:48.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 15:48:48.372 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:48:48.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:49:56.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 15:49:56.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 15:51:49.810 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 15:51:51.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 15:53:55.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.37s (2.06min) +2026-05-30 15:53:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 15:55:59.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:03:15.800 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 79, train_steps: 3160 +2026-05-30 16:03:16.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 16:03:17.024 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:03:17.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:04:25.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:04:25.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:06:19.676 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 16:06:20.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:08:26.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.06s (2.08min) +2026-05-30 16:08:27.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:10:33.727 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:17:51.706 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 80, train_steps: 3200 +2026-05-30 16:18:43.412 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-05-30 16:18:48.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 16:18:48.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:18:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:20:01.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:20:01.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:22:04.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 16:22:05.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:24:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.36s (2.07min) +2026-05-30 16:24:11.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:26:15.344 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:33:32.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 81, train_steps: 3240 +2026-05-30 16:33:33.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 16:33:34.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:33:34.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:34:42.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:34:42.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:36:39.523 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 16:36:40.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:38:45.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.43s (2.07min) +2026-05-30 16:38:46.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:40:58.782 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 16:48:14.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 82, train_steps: 3280 +2026-05-30 16:48:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 16:48:15.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:48:15.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:49:23.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 16:49:23.816 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 16:51:15.703 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 16:51:16.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 16:53:21.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.03s (2.07min) +2026-05-30 16:53:22.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 16:55:31.039 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:02:53.799 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 83, train_steps: 3320 +2026-05-30 17:02:55.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 17:02:55.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:02:55.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:04:03.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:04:03.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:05:56.257 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 17:05:57.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:08:05.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-30 17:08:06.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:10:14.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:17:33.752 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 84, train_steps: 3360 +2026-05-30 17:17:35.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 17:17:35.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:17:35.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:18:44.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:18:44.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:20:37.347 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 17:20:38.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:22:41.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.95s (2.05min) +2026-05-30 17:22:42.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:24:51.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:32:09.208 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 85, train_steps: 3400 +2026-05-30 17:32:10.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 17:32:10.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:32:10.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:33:19.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:33:19.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:35:25.656 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 17:35:26.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:37:30.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.82s (2.06min) +2026-05-30 17:37:31.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:39:34.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 17:46:53.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 86, train_steps: 3440 +2026-05-30 17:46:54.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 17:46:54.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:46:54.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:48:02.936 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 17:48:02.941 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 17:50:01.301 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 17:50:02.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 17:52:07.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.23s (2.07min) +2026-05-30 17:52:07.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 17:54:13.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:01:33.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 87, train_steps: 3480 +2026-05-30 18:01:35.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 18:01:35.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:01:35.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:02:43.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:02:43.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:04:39.602 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 18:04:40.650 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:06:46.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.99s (2.08min) +2026-05-30 18:06:47.303 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:08:57.840 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:16:16.544 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 88, train_steps: 3520 +2026-05-30 18:16:17.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 18:16:18.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:16:18.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:17:27.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:17:27.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:19:30.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 18:19:32.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:21:39.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.00s (2.12min) +2026-05-30 18:21:40.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:23:55.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:31:17.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 89, train_steps: 3560 +2026-05-30 18:31:18.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 18:31:18.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:31:18.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:32:29.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:32:29.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:34:45.309 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 18:34:46.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:36:52.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.24s (2.09min) +2026-05-30 18:36:53.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:39:01.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 18:46:22.140 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 90, train_steps: 3600 +2026-05-30 18:47:18.253 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-05-30 18:47:23.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 18:47:24.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:47:24.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:48:34.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 18:48:34.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 18:50:38.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 18:50:39.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 18:52:44.634 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.28s (2.07min) +2026-05-30 18:52:45.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 18:54:52.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:02:12.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 91, train_steps: 3640 +2026-05-30 19:02:13.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 19:02:13.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:02:13.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:03:25.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:03:25.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:05:33.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 19:05:34.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:07:42.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.27s (2.12min) +2026-05-30 19:07:43.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:09:52.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:17:11.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 92, train_steps: 3680 +2026-05-30 19:17:12.646 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 19:17:12.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:17:12.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:18:21.641 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:18:21.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:20:25.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 19:20:26.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:22:30.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.31s (2.06min) +2026-05-30 19:22:31.481 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:24:34.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:31:54.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 93, train_steps: 3720 +2026-05-30 19:31:55.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 19:31:55.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:31:55.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:33:05.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:33:05.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:35:23.054 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 19:35:24.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:37:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.42s (2.12min) +2026-05-30 19:37:33.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:39:42.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 19:47:01.992 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 94, train_steps: 3760 +2026-05-30 19:47:03.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 19:47:03.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:47:03.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:48:12.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 19:48:12.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 19:50:10.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 19:50:11.605 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 19:52:15.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.62s (2.06min) +2026-05-30 19:52:16.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 19:54:22.364 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:01:44.989 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 95, train_steps: 3800 +2026-05-30 20:01:46.251 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 20:01:46.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:01:46.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:02:56.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:02:56.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:04:55.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 20:04:56.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:07:03.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.05s (2.12min) +2026-05-30 20:07:04.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:09:09.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:16:28.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 96, train_steps: 3840 +2026-05-30 20:16:29.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 20:16:29.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:16:29.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:17:40.567 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:17:40.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:19:46.394 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 20:19:47.301 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:21:51.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.85s (2.06min) +2026-05-30 20:21:52.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:23:59.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:31:18.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 97, train_steps: 3880 +2026-05-30 20:31:19.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 20:31:20.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:31:20.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:32:28.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:32:28.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:34:23.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 20:34:24.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:36:33.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 128.11s (2.14min) +2026-05-30 20:36:34.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:38:40.356 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 20:46:01.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 98, train_steps: 3920 +2026-05-30 20:46:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 20:46:03.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:46:03.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:47:12.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 20:47:12.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 20:49:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 20:49:15.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 20:51:18.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.74s (2.05min) +2026-05-30 20:51:19.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 20:53:25.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:00:46.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 99, train_steps: 3960 +2026-05-30 21:00:47.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 21:00:47.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:00:47.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:01:55.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:01:55.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:03:57.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 21:03:58.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:06:03.933 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.03s (2.08min) +2026-05-30 21:06:04.852 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:08:07.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:15:25.181 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 100, train_steps: 4000 +2026-05-30 21:17:25.342 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-05-30 21:17:31.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 21:17:31.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:17:31.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:18:40.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:18:40.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:20:36.958 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 21:20:37.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 21:22:45.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.62s (2.11min) +2026-05-30 21:22:46.115 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 21:24:51.442 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 21:32:07.584 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 101, train_steps: 4040 +2026-05-30 21:32:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 21:32:09.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:32:09.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:48:43.867 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-30 21:48:43.869 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-30 21:55:22.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 21:55:22.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:55:22.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:56:47.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 21:56:47.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 21:58:52.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 21:59:08.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:01:15.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.76s (2.11min) +2026-05-30 22:01:16.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:03:17.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:03:22.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2806 - Rank 9, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-05-30 22:10:51.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 1, train_steps: 40 +2026-05-30 22:10:52.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 22:10:52.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:10:52.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:11:59.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:11:59.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:14:02.396 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 22:14:03.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:16:11.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.28s (2.12min) +2026-05-30 22:16:12.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:18:16.877 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:25:32.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 2, train_steps: 80 +2026-05-30 22:25:33.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 22:25:33.514 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:25:33.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:26:41.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:26:41.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:28:36.345 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 22:28:37.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:30:39.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.15s (2.04min) +2026-05-30 22:30:40.807 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:32:47.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:40:02.636 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 3, train_steps: 120 +2026-05-30 22:40:03.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 22:40:03.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:40:03.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:41:11.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:41:11.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:43:09.821 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 22:43:10.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:45:16.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.11s (2.09min) +2026-05-30 22:45:17.402 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 22:47:24.527 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 22:54:37.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 4, train_steps: 160 +2026-05-30 22:54:38.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 22:54:38.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:54:39.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:55:46.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 22:55:46.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 22:57:49.071 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 22:57:50.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 22:59:58.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.78s (2.13min) +2026-05-30 22:59:59.448 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:02:05.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:09:18.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 5, train_steps: 200 +2026-05-30 23:09:20.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 23:09:20.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:09:20.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:10:27.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:10:27.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:12:35.172 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 23:12:36.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:14:47.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 131.27s (2.19min) +2026-05-30 23:14:48.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:16:52.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:24:04.246 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 6, train_steps: 240 +2026-05-30 23:24:05.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 23:24:05.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:24:05.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:25:13.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:25:13.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:27:13.844 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 23:27:14.773 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:29:20.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.98s (2.08min) +2026-05-30 23:29:21.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:31:29.607 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:38:45.285 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 7, train_steps: 280 +2026-05-30 23:38:46.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 23:38:46.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:38:46.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:39:54.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:39:54.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:41:47.769 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 23:41:48.625 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:43:52.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.38s (2.06min) +2026-05-30 23:43:53.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-30 23:45:54.759 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-30 23:53:08.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 8, train_steps: 320 +2026-05-30 23:53:10.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-30 23:53:10.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:53:10.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:54:17.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-30 23:54:17.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-30 23:56:14.870 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-30 23:56:15.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-30 23:58:20.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.49s (2.07min) +2026-05-30 23:58:21.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:00:25.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:07:40.423 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 9, train_steps: 360 +2026-05-31 00:07:41.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 00:07:41.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:07:41.649 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:08:53.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:08:53.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:10:45.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 00:10:46.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:12:48.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 122.20s (2.04min) +2026-05-31 00:12:49.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:14:52.721 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:22:04.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 10, train_steps: 400 +2026-05-31 00:22:54.210 | INFO | __main__:maybe_update_ref_model_for_kl:797 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-05-31 00:22:59.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 00:22:59.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:22:59.932 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:24:08.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:24:08.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:26:04.324 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 00:26:05.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:28:11.601 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 125.43s (2.09min) +2026-05-31 00:28:12.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:30:15.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:37:29.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 11, train_steps: 440 +2026-05-31 00:37:30.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 00:37:30.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:37:30.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:38:39.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:38:39.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:40:45.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 00:40:46.321 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:42:50.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.47s (2.06min) +2026-05-31 00:42:51.045 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:44:54.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 00:52:06.312 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 12, train_steps: 480 +2026-05-31 00:52:07.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 00:52:07.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:52:07.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:53:14.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 00:53:14.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 00:55:21.080 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 00:55:22.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 00:57:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.24s (2.05min) +2026-05-31 00:57:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 00:59:29.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:06:43.389 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 13, train_steps: 520 +2026-05-31 01:06:44.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 01:06:44.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:06:44.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:07:53.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:07:53.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:09:51.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 01:09:52.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:11:59.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 126.39s (2.11min) +2026-05-31 01:12:00.297 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:14:01.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:21:15.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 14, train_steps: 560 +2026-05-31 01:21:16.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 01:21:16.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:21:17.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:22:23.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:22:23.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:24:19.346 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 01:24:20.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:26:19.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 118.28s (1.97min) +2026-05-31 01:26:20.374 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:28:23.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:35:35.563 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 15, train_steps: 600 +2026-05-31 01:35:36.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 01:35:36.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:35:36.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:36:44.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:36:44.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:38:37.372 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 01:38:38.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:40:43.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.29s (2.07min) +2026-05-31 01:40:44.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:42:47.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 01:50:02.595 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 16, train_steps: 640 +2026-05-31 01:50:03.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 01:50:03.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:50:03.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:51:10.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 01:51:10.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 01:53:05.978 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 01:53:06.815 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 01:55:15.387 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 127.91s (2.13min) +2026-05-31 01:55:16.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 01:57:21.330 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:04:32.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 17, train_steps: 680 +2026-05-31 02:04:33.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 02:04:34.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:04:34.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:05:42.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:05:42.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:07:38.765 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 02:07:39.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:09:44.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.66s (2.08min) +2026-05-31 02:09:45.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:11:47.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-05-31 02:19:00.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3244 - Rank 9, sp_rank 0: update step: 18, train_steps: 720 +2026-05-31 02:19:01.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-05-31 02:19:01.261 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:19:01.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:20:08.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-31 02:20:08.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-05-31 02:22:01.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-05-31 02:22:02.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-05-31 02:24:07.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 123.93s (2.07min) +2026-05-31 02:24:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-31 02:26:11.757 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:00:37.837 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-06-25 18:00:37.838 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-06-25 18:04:09.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 18:04:09.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:04:09.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:05:30.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:05:30.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:07:16.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 18:07:32.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:09:27.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.93s (1.92min) +2026-06-25 18:09:28.015 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:11:24.298 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:11:27.504 | INFO | hyvideo.utils.grpo_utils:train_one_step:3127 - Rank 9, sp_rank 0: init adaptive KL controller (init=1.000e-07, min=1.000e-07, max=1.000e-05) +2026-06-25 18:18:40.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 1, train_steps: 40 +2026-06-25 18:18:41.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 18:18:41.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:18:41.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:19:46.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:19:46.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:21:34.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 18:21:35.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:23:29.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.84s (1.90min) +2026-06-25 18:23:29.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:25:26.712 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:32:25.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 2, train_steps: 80 +2026-06-25 18:32:26.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 18:32:26.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:32:26.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:33:31.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:33:31.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:35:14.571 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 18:35:15.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:37:08.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.32s (1.89min) +2026-06-25 18:37:09.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:39:06.980 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 18:46:07.315 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 3, train_steps: 120 +2026-06-25 18:46:07.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 18:46:08.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:46:08.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:47:13.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 18:47:13.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 18:49:02.781 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 18:49:03.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 18:50:58.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.18s (1.90min) +2026-06-25 18:50:58.938 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 18:52:58.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:00:02.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 4, train_steps: 160 +2026-06-25 19:00:02.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 19:00:02.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:00:03.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:01:08.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:01:08.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:02:56.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 19:02:57.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:04:51.964 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.26s (1.90min) +2026-06-25 19:04:52.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:06:49.822 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:13:48.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 5, train_steps: 200 +2026-06-25 19:13:49.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 19:13:49.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:13:49.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:14:54.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:14:54.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:16:41.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 19:16:42.530 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:18:35.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-25 19:18:35.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:20:32.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:27:31.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 6, train_steps: 240 +2026-06-25 19:27:32.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 19:27:32.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:27:32.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:28:37.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:28:37.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:30:22.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 19:30:24.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:32:17.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.74s (1.88min) +2026-06-25 19:32:17.756 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:34:15.139 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:41:16.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 7, train_steps: 280 +2026-06-25 19:41:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 19:41:17.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:41:17.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:42:22.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:42:22.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:44:14.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 19:44:15.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:46:09.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.77s (1.90min) +2026-06-25 19:46:10.187 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 19:48:07.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 19:55:07.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 8, train_steps: 320 +2026-06-25 19:55:08.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 19:55:08.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:55:08.795 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:56:14.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 19:56:14.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 19:58:02.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 19:58:03.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 19:59:57.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-25 19:59:58.359 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:01:58.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:09:01.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 9, train_steps: 360 +2026-06-25 20:09:02.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 20:09:02.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:09:02.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:10:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:10:07.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:11:59.728 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 20:12:00.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:13:55.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.44s (1.91min) +2026-06-25 20:13:55.998 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:15:53.500 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:22:53.803 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 10, train_steps: 400 +2026-06-25 20:23:30.097 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=10 (missing=0, unexpected=0) +2026-06-25 20:23:32.341 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 20:23:32.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:23:32.556 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:24:39.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:24:39.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:26:25.325 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 20:26:26.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:28:19.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-25 20:28:19.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:30:18.624 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:37:18.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 11, train_steps: 440 +2026-06-25 20:37:18.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 20:37:18.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:37:19.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:38:24.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:38:24.026 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:40:08.461 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 20:40:09.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:42:02.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.92s (1.88min) +2026-06-25 20:42:03.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:44:00.747 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 20:51:02.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 12, train_steps: 480 +2026-06-25 20:51:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 20:51:03.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:51:03.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:52:08.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 20:52:08.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 20:53:56.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 20:53:57.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 20:55:51.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.41s (1.91min) +2026-06-25 20:55:52.296 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 20:57:49.280 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:04:50.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 13, train_steps: 520 +2026-06-25 21:04:50.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 21:04:50.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:04:50.954 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:05:56.065 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:05:56.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:07:48.686 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 21:07:49.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:09:42.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-25 21:09:43.496 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:11:42.031 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:18:43.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 14, train_steps: 560 +2026-06-25 21:18:43.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 21:18:43.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:18:43.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:19:48.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:19:48.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:21:38.761 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 21:21:39.744 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:23:32.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.64s (1.88min) +2026-06-25 21:23:33.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:25:30.191 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:32:30.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 15, train_steps: 600 +2026-06-25 21:32:31.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 21:32:31.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:32:31.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:33:36.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:33:36.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:35:20.629 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 21:35:21.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:37:15.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.70s (1.90min) +2026-06-25 21:37:16.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:39:14.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 21:46:15.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 16, train_steps: 640 +2026-06-25 21:46:16.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 21:46:16.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:46:16.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:47:21.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 21:47:21.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 21:49:09.537 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 21:49:10.764 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 21:51:03.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.68s (1.88min) +2026-06-25 21:51:04.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 21:53:01.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:00:05.050 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 17, train_steps: 680 +2026-06-25 22:00:05.671 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 22:00:05.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:00:05.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:01:11.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:01:11.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:02:57.753 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 22:02:58.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:04:53.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.49s (1.91min) +2026-06-25 22:04:54.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:06:50.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:13:49.814 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 18, train_steps: 720 +2026-06-25 22:13:50.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 22:13:50.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:13:50.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:14:55.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:14:55.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:16:44.673 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 22:16:45.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:18:38.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.63s (1.88min) +2026-06-25 22:18:39.102 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:20:37.931 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:27:38.748 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 19, train_steps: 760 +2026-06-25 22:27:39.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 22:27:39.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:27:39.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:28:44.897 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:28:44.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:30:31.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 22:30:32.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:32:25.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.91s (1.88min) +2026-06-25 22:32:25.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:34:23.599 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:41:22.091 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 20, train_steps: 800 +2026-06-25 22:41:57.988 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=20 (missing=0, unexpected=0) +2026-06-25 22:41:59.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 22:41:59.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:41:59.975 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:43:06.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:43:06.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:44:51.320 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 22:44:52.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 22:46:44.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.30s (1.87min) +2026-06-25 22:46:45.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 22:48:44.065 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 22:55:44.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 21, train_steps: 840 +2026-06-25 22:55:45.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 22:55:45.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:55:45.754 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:56:50.889 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 22:56:50.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 22:58:38.405 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 22:58:39.298 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:00:32.353 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-25 23:00:33.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:02:30.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:09:29.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 22, train_steps: 880 +2026-06-25 23:09:30.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 23:09:30.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:09:30.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:10:35.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:10:35.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:12:21.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 23:12:22.798 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:14:15.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.95s (1.88min) +2026-06-25 23:14:16.694 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:16:15.107 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:23:14.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 23, train_steps: 920 +2026-06-25 23:23:14.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 23:23:15.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:23:15.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:24:20.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:24:20.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:26:10.058 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 23:26:10.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:28:05.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-25 23:28:05.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:30:04.871 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:37:06.361 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 24, train_steps: 960 +2026-06-25 23:37:07.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 23:37:07.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:37:07.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:38:12.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:38:12.354 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:40:00.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 23:40:02.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:41:56.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-25 23:41:56.923 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:43:54.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-25 23:50:53.555 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 25, train_steps: 1000 +2026-06-25 23:51:51.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-25 23:51:52.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:51:52.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:52:57.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-25 23:52:57.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-25 23:54:42.933 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-25 23:54:43.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-25 23:56:36.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.76s (1.88min) +2026-06-25 23:56:37.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-25 23:58:36.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:05:35.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 26, train_steps: 1040 +2026-06-26 00:05:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 00:05:36.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:05:36.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:06:41.661 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:06:41.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:08:27.266 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 00:08:28.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:10:21.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-26 00:10:22.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:12:20.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:19:20.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 27, train_steps: 1080 +2026-06-26 00:19:21.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 00:19:21.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:19:21.386 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:20:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:20:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:22:10.367 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 00:22:11.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:24:04.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 00:24:05.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:26:05.222 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:33:06.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 28, train_steps: 1120 +2026-06-26 00:33:07.428 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 00:33:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:33:07.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:34:12.906 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:34:12.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:36:02.990 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 00:36:04.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:37:58.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.22s (1.90min) +2026-06-26 00:37:59.275 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:39:59.272 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 00:47:02.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 29, train_steps: 1160 +2026-06-26 00:47:03.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 00:47:03.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:47:03.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:48:08.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 00:48:08.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 00:49:59.961 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 00:50:01.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 00:51:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.60s (1.89min) +2026-06-26 00:51:55.433 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 00:53:52.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:00:52.528 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 30, train_steps: 1200 +2026-06-26 01:01:28.716 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=30 (missing=0, unexpected=0) +2026-06-26 01:01:30.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 01:01:30.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:01:30.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:02:37.551 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:02:37.555 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:04:24.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 01:04:25.849 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:06:18.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:06:19.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:08:18.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:15:17.114 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 31, train_steps: 1240 +2026-06-26 01:15:17.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 01:15:18.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:15:18.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:16:23.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:16:23.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:18:11.059 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 01:18:12.014 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:20:06.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.00s (1.90min) +2026-06-26 01:20:06.849 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:22:06.592 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:29:07.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 32, train_steps: 1280 +2026-06-26 01:29:08.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 01:29:08.695 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:29:08.800 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:30:13.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:30:13.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:32:01.905 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 01:32:03.180 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:33:57.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.16s (1.90min) +2026-06-26 01:33:58.159 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:35:57.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:42:57.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 33, train_steps: 1320 +2026-06-26 01:42:58.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 01:42:58.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:42:58.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:44:03.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:44:03.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:45:49.639 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 01:45:50.599 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 01:47:43.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.47s (1.87min) +2026-06-26 01:47:43.867 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 01:49:43.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 01:56:43.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 34, train_steps: 1360 +2026-06-26 01:56:44.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 01:56:44.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:56:44.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:57:49.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 01:57:49.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 01:59:39.166 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 01:59:40.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:01:33.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 02:01:33.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:03:30.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:10:30.221 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 35, train_steps: 1400 +2026-06-26 02:10:30.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 02:10:31.005 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:10:31.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:11:36.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:11:36.738 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:13:23.743 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 02:13:24.981 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:15:17.703 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 02:15:18.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:17:15.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:24:15.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 36, train_steps: 1440 +2026-06-26 02:24:15.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 02:24:16.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:24:16.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:25:21.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:25:21.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:27:10.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 02:27:11.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:29:06.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.29s (1.90min) +2026-06-26 02:29:06.990 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:31:06.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:38:07.252 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 37, train_steps: 1480 +2026-06-26 02:38:07.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 02:38:08.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:38:08.242 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:39:13.271 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:39:13.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:40:59.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 02:41:00.078 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:42:54.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.15s (1.90min) +2026-06-26 02:42:55.044 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:44:52.413 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 02:51:53.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 38, train_steps: 1520 +2026-06-26 02:51:53.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 02:51:53.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:51:53.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:52:59.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 02:52:59.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 02:54:47.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 02:54:48.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 02:56:41.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.87s (1.88min) +2026-06-26 02:56:42.103 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 02:58:39.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:05:39.950 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 39, train_steps: 1560 +2026-06-26 03:05:40.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 03:05:40.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:05:40.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:06:46.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:06:46.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:08:37.515 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 03:08:38.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:10:31.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.21s (1.89min) +2026-06-26 03:10:32.507 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:12:29.843 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:19:28.205 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 40, train_steps: 1600 +2026-06-26 03:20:05.010 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=40 (missing=0, unexpected=0) +2026-06-26 03:20:06.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 03:20:06.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:20:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:21:13.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:21:13.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:22:58.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 03:22:59.542 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:24:53.857 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.23s (1.90min) +2026-06-26 03:24:54.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:26:52.110 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:33:51.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 41, train_steps: 1640 +2026-06-26 03:33:52.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 03:33:52.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:33:52.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:34:58.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:34:58.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:36:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 03:36:46.352 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:38:38.948 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:38:39.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:40:38.716 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 03:47:39.336 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 42, train_steps: 1680 +2026-06-26 03:47:40.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 03:47:40.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:47:40.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:48:45.457 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 03:48:45.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 03:50:30.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 03:50:31.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 03:52:23.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.45s (1.87min) +2026-06-26 03:52:24.419 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 03:54:21.286 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:01:20.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 43, train_steps: 1720 +2026-06-26 04:01:21.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 04:01:21.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:01:21.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:02:26.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:02:26.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:04:11.575 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 04:04:12.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:06:06.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 04:06:07.052 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:08:06.619 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:15:08.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 44, train_steps: 1760 +2026-06-26 04:15:09.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 04:15:09.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:15:09.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:16:15.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:16:15.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:18:04.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 04:18:05.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:19:59.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.96s (1.90min) +2026-06-26 04:20:00.517 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:22:00.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:29:02.160 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 45, train_steps: 1800 +2026-06-26 04:29:02.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 04:29:03.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:29:03.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:30:08.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:30:08.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:31:53.168 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 04:31:54.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:33:46.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 04:33:47.302 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:35:44.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:42:45.606 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 46, train_steps: 1840 +2026-06-26 04:42:46.330 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 04:42:46.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:42:46.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:43:51.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:43:51.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:45:34.830 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 04:45:35.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 04:47:29.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.41s (1.89min) +2026-06-26 04:47:30.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 04:49:27.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 04:56:26.882 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 47, train_steps: 1880 +2026-06-26 04:56:27.578 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 04:56:27.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:56:27.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:57:33.092 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 04:57:33.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 04:59:18.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 04:59:19.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:01:12.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 05:01:13.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:03:10.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:10:09.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 48, train_steps: 1920 +2026-06-26 05:10:10.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 05:10:10.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:10:10.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:11:15.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:11:15.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:13:02.653 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 05:13:03.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:14:58.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.90s (1.90min) +2026-06-26 05:14:58.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:16:58.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:24:02.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 49, train_steps: 1960 +2026-06-26 05:24:02.895 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 05:24:03.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:24:03.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:25:08.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:25:08.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:26:57.937 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 05:26:58.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:28:52.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.89s (1.90min) +2026-06-26 05:28:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:30:50.660 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:37:49.571 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 50, train_steps: 2000 +2026-06-26 05:39:26.099 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=50 (missing=0, unexpected=0) +2026-06-26 05:39:28.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 05:39:29.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:39:29.152 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:40:35.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:40:35.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:42:19.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 05:42:20.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:44:12.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.10s (1.87min) +2026-06-26 05:44:13.278 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:46:10.576 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 05:53:09.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 51, train_steps: 2040 +2026-06-26 05:53:10.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 05:53:10.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:53:10.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:54:16.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 05:54:16.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 05:55:59.884 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 05:56:00.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 05:57:53.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.80s (1.88min) +2026-06-26 05:57:54.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 05:59:51.601 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:06:51.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 52, train_steps: 2080 +2026-06-26 06:06:52.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 06:06:52.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:06:52.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:07:57.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:07:57.592 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:09:43.925 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 06:09:44.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:11:38.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 06:11:38.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:13:38.166 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:20:39.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 53, train_steps: 2120 +2026-06-26 06:20:39.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 06:20:40.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:20:40.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:21:45.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:21:45.256 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:23:30.435 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 06:23:31.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:25:23.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.26s (1.87min) +2026-06-26 06:25:24.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:27:21.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:34:21.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 54, train_steps: 2160 +2026-06-26 06:34:21.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 06:34:21.912 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:34:21.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:35:27.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:35:27.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:37:10.500 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 06:37:11.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:39:06.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.48s (1.91min) +2026-06-26 06:39:06.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:41:06.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 06:48:08.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 55, train_steps: 2200 +2026-06-26 06:48:09.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 06:48:09.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:48:09.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:49:14.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 06:49:14.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 06:51:02.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 06:51:03.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 06:52:58.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.19s (1.90min) +2026-06-26 06:52:58.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 06:54:58.639 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:02:02.276 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 56, train_steps: 2240 +2026-06-26 07:02:03.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 07:02:03.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:02:03.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:03:08.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:03:08.702 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:04:54.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 07:04:55.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:06:49.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.50s (1.91min) +2026-06-26 07:06:50.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:08:47.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:15:47.796 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 57, train_steps: 2280 +2026-06-26 07:15:48.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 07:15:48.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:15:48.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:16:53.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:16:53.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:18:40.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 07:18:41.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:20:34.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.58s (1.88min) +2026-06-26 07:20:35.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:22:32.421 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:29:31.362 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 58, train_steps: 2320 +2026-06-26 07:29:32.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 07:29:32.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:29:32.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:30:37.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:30:37.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:32:30.098 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 07:32:30.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:34:23.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 07:34:24.456 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:36:21.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:43:20.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 59, train_steps: 2360 +2026-06-26 07:43:21.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 07:43:21.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:43:21.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:44:26.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:44:26.727 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:46:14.527 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 07:46:15.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 07:48:09.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.66s (1.89min) +2026-06-26 07:48:09.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 07:50:07.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 07:57:07.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 60, train_steps: 2400 +2026-06-26 07:57:43.941 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=60 (missing=0, unexpected=0) +2026-06-26 07:57:45.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 07:57:45.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:57:46.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 07:58:52.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 07:58:52.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:00:39.514 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 08:00:40.676 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:02:33.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-26 08:02:34.538 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:04:32.199 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:11:32.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 61, train_steps: 2440 +2026-06-26 08:11:33.149 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 08:11:33.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:11:33.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:12:38.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:12:38.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:14:25.510 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 08:14:26.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:16:18.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 08:16:19.690 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:18:18.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:25:17.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 62, train_steps: 2480 +2026-06-26 08:25:18.317 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 08:25:18.452 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:25:18.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:26:23.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:26:23.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:28:07.042 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 08:28:07.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:30:00.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 08:30:01.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:32:00.731 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:39:02.661 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 63, train_steps: 2520 +2026-06-26 08:39:03.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 08:39:03.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:39:03.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:40:08.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:40:08.883 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:41:56.217 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 08:41:57.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:43:51.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.64s (1.91min) +2026-06-26 08:43:52.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:45:49.788 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 08:52:49.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 64, train_steps: 2560 +2026-06-26 08:52:49.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 08:52:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:52:50.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:53:55.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 08:53:55.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 08:55:44.235 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 08:55:45.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 08:57:38.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 08:57:38.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 08:59:38.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:06:38.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 65, train_steps: 2600 +2026-06-26 09:06:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 09:06:39.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:06:39.444 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:07:44.582 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:07:44.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:09:32.385 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 09:09:33.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:11:25.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 09:11:26.322 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:13:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:20:21.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 66, train_steps: 2640 +2026-06-26 09:20:22.682 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 09:20:22.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:20:22.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:21:27.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:21:27.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:23:11.827 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 09:23:12.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:25:06.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-26 09:25:07.692 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:27:07.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:34:08.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 67, train_steps: 2680 +2026-06-26 09:34:09.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 09:34:09.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:34:09.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:35:14.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:35:14.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:37:03.160 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 09:37:04.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:38:57.218 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 09:38:57.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:40:56.077 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 09:47:56.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 68, train_steps: 2720 +2026-06-26 09:47:56.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 09:47:56.863 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:47:56.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:49:02.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 09:49:02.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 09:50:50.362 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 09:50:51.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 09:52:43.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.54s (1.88min) +2026-06-26 09:52:44.676 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 09:54:43.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:01:44.460 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 69, train_steps: 2760 +2026-06-26 10:01:45.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 10:01:45.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:01:45.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:02:50.502 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:02:50.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:04:34.030 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 10:04:34.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:06:27.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 10:06:27.760 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:08:24.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:15:23.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 70, train_steps: 2800 +2026-06-26 10:15:59.613 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=70 (missing=0, unexpected=0) +2026-06-26 10:16:01.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 10:16:01.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:16:02.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:17:08.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:17:08.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:18:58.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 10:18:58.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:20:53.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.37s (1.91min) +2026-06-26 10:20:54.273 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:22:51.683 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:29:51.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 71, train_steps: 2840 +2026-06-26 10:29:52.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 10:29:52.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:29:52.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:30:57.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:30:57.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:32:43.317 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 10:32:44.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:34:37.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 10:34:37.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:36:37.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:43:36.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 72, train_steps: 2880 +2026-06-26 10:43:37.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 10:43:37.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:43:37.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:44:42.662 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:44:42.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:46:28.282 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 10:46:29.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 10:48:22.133 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 10:48:22.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 10:50:20.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 10:57:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 73, train_steps: 2920 +2026-06-26 10:57:21.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 10:57:21.339 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:57:21.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 10:58:26.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 10:58:26.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:00:16.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 11:00:18.208 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:02:10.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.38s (1.87min) +2026-06-26 11:02:11.511 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:04:08.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:11:08.051 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 74, train_steps: 2960 +2026-06-26 11:11:08.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 11:11:08.923 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:11:08.984 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:12:13.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:12:13.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:13:58.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 11:13:59.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:15:52.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 11:15:53.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:17:50.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:24:50.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 75, train_steps: 3000 +2026-06-26 11:26:08.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 11:26:08.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:26:08.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:27:13.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:27:13.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:29:02.143 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 11:29:03.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:30:56.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-26 11:30:56.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:32:54.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:39:54.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 76, train_steps: 3040 +2026-06-26 11:39:54.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 11:39:55.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:39:55.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:41:00.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:41:00.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:42:43.574 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 11:42:44.471 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:44:37.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.36s (1.87min) +2026-06-26 11:44:37.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 11:46:36.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 11:53:36.197 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 77, train_steps: 3080 +2026-06-26 11:53:36.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 11:53:37.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:53:37.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:54:42.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 11:54:42.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 11:56:30.211 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 11:56:31.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 11:58:24.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 11:58:24.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:00:21.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:07:20.510 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 78, train_steps: 3120 +2026-06-26 12:07:21.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 12:07:21.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:07:21.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:08:26.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:08:26.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:10:11.287 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 12:10:12.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:12:05.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.61s (1.89min) +2026-06-26 12:12:06.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:14:06.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:21:06.000 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 79, train_steps: 3160 +2026-06-26 12:21:06.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 12:21:06.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:21:07.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:22:12.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:22:12.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:23:59.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 12:24:00.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:25:53.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 12:25:54.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:27:51.926 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:34:52.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 80, train_steps: 3200 +2026-06-26 12:35:29.005 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=80 (missing=0, unexpected=0) +2026-06-26 12:35:30.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 12:35:30.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:35:30.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:36:38.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:36:38.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:38:24.341 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 12:38:25.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:40:18.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-26 12:40:18.961 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:42:17.260 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 12:49:15.425 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 81, train_steps: 3240 +2026-06-26 12:49:16.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 12:49:16.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:49:16.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:50:21.050 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 12:50:21.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 12:52:08.745 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 12:52:09.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 12:54:03.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 12:54:03.808 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 12:56:00.936 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:03:02.080 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 82, train_steps: 3280 +2026-06-26 13:03:02.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 13:03:03.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:03:03.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:04:08.768 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:04:08.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:05:52.210 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 13:05:53.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:07:46.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 13:07:46.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:09:44.098 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:16:43.230 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 83, train_steps: 3320 +2026-06-26 13:16:43.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 13:16:44.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:16:44.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:17:49.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:17:49.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:19:40.622 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 13:19:41.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:21:34.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.72s (1.88min) +2026-06-26 13:21:35.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:23:32.739 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:30:30.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 84, train_steps: 3360 +2026-06-26 13:30:31.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 13:30:31.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:30:31.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:31:36.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:31:36.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:33:24.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 13:33:25.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:35:18.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-26 13:35:18.726 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:37:16.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:44:14.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 85, train_steps: 3400 +2026-06-26 13:44:15.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 13:44:15.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:44:15.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:45:20.456 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:45:20.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:47:12.702 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 13:47:13.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 13:49:07.573 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.79s (1.90min) +2026-06-26 13:49:08.283 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 13:51:06.687 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 13:58:06.627 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 86, train_steps: 3440 +2026-06-26 13:58:07.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 13:58:07.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:58:07.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 13:59:12.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 13:59:12.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:00:55.999 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 14:00:56.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:02:51.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.45s (1.91min) +2026-06-26 14:02:52.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:04:50.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:11:48.568 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 87, train_steps: 3480 +2026-06-26 14:11:49.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 14:11:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:11:49.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:12:54.808 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:12:54.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:14:38.146 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 14:14:39.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:16:32.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-26 14:16:32.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:18:29.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:25:27.183 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 88, train_steps: 3520 +2026-06-26 14:25:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 14:25:28.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:25:28.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:26:33.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:26:33.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:28:24.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 14:28:25.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:30:17.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.59s (1.88min) +2026-06-26 14:30:18.664 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:32:16.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:39:14.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 89, train_steps: 3560 +2026-06-26 14:39:15.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 14:39:15.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:39:15.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:40:21.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:40:21.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:42:12.102 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 14:42:13.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:44:07.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.01s (1.90min) +2026-06-26 14:44:07.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 14:46:06.386 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 14:53:06.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 90, train_steps: 3600 +2026-06-26 14:53:42.186 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=90 (missing=0, unexpected=0) +2026-06-26 14:53:45.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 14:53:45.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:53:45.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:54:52.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 14:54:52.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 14:56:35.578 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 14:56:36.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 14:58:29.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.12s (1.87min) +2026-06-26 14:58:29.773 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:00:27.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:07:26.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 91, train_steps: 3640 +2026-06-26 15:07:27.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 15:07:27.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:07:27.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:08:32.839 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:08:32.842 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:10:18.939 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 15:10:19.915 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:12:12.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.98s (1.87min) +2026-06-26 15:12:12.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:14:10.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:21:10.133 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 92, train_steps: 3680 +2026-06-26 15:21:10.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 15:21:11.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:21:11.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:22:16.583 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:22:16.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:24:01.112 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 15:24:02.496 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:25:55.408 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.86s (1.88min) +2026-06-26 15:25:56.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:27:53.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:34:55.257 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 93, train_steps: 3720 +2026-06-26 15:34:55.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 15:34:56.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:34:56.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:36:01.571 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:36:01.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:37:49.469 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 15:37:50.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:39:43.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.94s (1.88min) +2026-06-26 15:39:44.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:41:43.265 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 15:48:44.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 94, train_steps: 3760 +2026-06-26 15:48:45.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 15:48:45.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:48:45.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:49:50.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 15:49:50.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 15:51:34.720 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 15:51:35.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 15:53:28.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.79s (1.88min) +2026-06-26 15:53:29.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 15:55:26.353 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:02:25.888 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 95, train_steps: 3800 +2026-06-26 16:02:26.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 16:02:26.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:02:26.899 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:03:32.060 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:03:32.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:05:16.922 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 16:05:17.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:07:10.362 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.22s (1.87min) +2026-06-26 16:07:11.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:09:08.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:16:06.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 96, train_steps: 3840 +2026-06-26 16:16:06.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 16:16:06.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:16:06.870 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:17:12.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:17:12.084 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:19:03.074 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 16:19:04.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:20:56.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.53s (1.88min) +2026-06-26 16:20:57.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:22:55.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:29:53.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 97, train_steps: 3880 +2026-06-26 16:29:54.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 16:29:54.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:29:54.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:31:00.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:31:00.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:32:44.931 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 16:32:45.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:34:38.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.29s (1.87min) +2026-06-26 16:34:39.120 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:36:38.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:43:37.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 98, train_steps: 3920 +2026-06-26 16:43:38.088 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 16:43:38.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:43:38.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:44:43.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:44:43.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:46:30.659 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 16:46:31.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 16:48:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-26 16:48:25.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 16:50:22.653 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 16:57:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 99, train_steps: 3960 +2026-06-26 16:57:20.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 16:57:20.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:57:21.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 16:58:26.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 16:58:26.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:00:14.957 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 17:00:15.931 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:02:09.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.36s (1.89min) +2026-06-26 17:02:10.104 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:04:07.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:11:06.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 100, train_steps: 4000 +2026-06-26 17:12:51.241 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=100 (missing=0, unexpected=0) +2026-06-26 17:12:52.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 17:12:53.077 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:12:53.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:13:59.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:13:59.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:15:47.056 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 17:15:47.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:17:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.20s (1.87min) +2026-06-26 17:17:41.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:19:39.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:26:38.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 101, train_steps: 4040 +2026-06-26 17:26:39.615 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 17:26:39.781 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:26:39.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:27:44.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:27:45.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:29:29.390 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 17:29:30.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:31:23.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.85s (1.88min) +2026-06-26 17:31:24.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:33:21.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:40:19.453 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 102, train_steps: 4080 +2026-06-26 17:40:20.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 17:40:20.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:40:20.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:41:25.663 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:41:25.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:43:13.843 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 17:43:14.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:45:08.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-26 17:45:09.178 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 17:47:07.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 17:54:05.949 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 103, train_steps: 4120 +2026-06-26 17:54:06.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 17:54:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:54:06.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:55:11.794 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 17:55:11.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 17:56:57.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 17:56:58.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 17:58:52.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.93s (1.90min) +2026-06-26 17:58:53.719 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:00:50.621 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:07:48.654 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 104, train_steps: 4160 +2026-06-26 18:07:49.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 18:07:49.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:07:49.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:08:54.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:08:54.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:10:44.981 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 18:10:45.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:12:38.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.14s (1.87min) +2026-06-26 18:12:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:14:38.306 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:21:37.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 105, train_steps: 4200 +2026-06-26 18:21:38.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 18:21:38.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:21:38.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:22:43.809 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:22:43.810 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:24:36.190 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 18:24:37.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:26:30.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-26 18:26:30.845 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:28:27.995 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:35:26.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 106, train_steps: 4240 +2026-06-26 18:35:27.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 18:35:27.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:35:28.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:36:33.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:36:33.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:38:23.638 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 18:38:24.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:40:17.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.69s (1.88min) +2026-06-26 18:40:18.401 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:42:15.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 18:49:14.304 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 107, train_steps: 4280 +2026-06-26 18:49:15.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 18:49:15.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:49:15.200 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:50:20.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 18:50:20.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 18:52:04.068 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 18:52:05.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 18:53:58.113 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.78s (1.88min) +2026-06-26 18:53:58.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 18:55:58.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:03:00.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 108, train_steps: 4320 +2026-06-26 19:03:01.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 19:03:01.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:03:01.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:04:07.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:04:07.877 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:05:54.689 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 19:05:55.595 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:07:48.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.15s (1.89min) +2026-06-26 19:07:49.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:09:46.749 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:16:45.609 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 109, train_steps: 4360 +2026-06-26 19:16:46.312 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 19:16:46.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:16:46.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:17:51.845 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:17:51.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:19:39.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 19:19:40.461 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:21:33.586 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.93s (1.88min) +2026-06-26 19:21:34.328 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:23:31.363 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:30:31.008 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 110, train_steps: 4400 +2026-06-26 19:31:06.798 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=110 (missing=0, unexpected=0) +2026-06-26 19:31:08.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 19:31:09.048 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:31:09.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:32:16.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:32:16.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:34:06.091 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 19:34:07.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:35:59.631 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-26 19:36:00.379 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:38:00.121 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:45:02.161 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 111, train_steps: 4440 +2026-06-26 19:45:02.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 19:45:03.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:45:03.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:46:08.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:46:08.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:47:51.716 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 19:47:52.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 19:49:44.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-26 19:49:45.697 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 19:51:44.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 19:58:45.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 112, train_steps: 4480 +2026-06-26 19:58:45.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 19:58:45.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:58:45.992 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 19:59:51.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 19:59:51.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:01:37.023 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 20:01:37.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:03:30.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.65s (1.88min) +2026-06-26 20:03:31.570 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:05:28.833 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:12:27.548 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 113, train_steps: 4520 +2026-06-26 20:12:28.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 20:12:28.419 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:12:28.510 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:13:33.489 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:13:33.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:15:21.129 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 20:15:22.023 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:17:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-26 20:17:15.167 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:19:14.741 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:26:13.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 114, train_steps: 4560 +2026-06-26 20:26:14.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 20:26:14.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:26:14.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:27:19.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:27:19.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:29:11.826 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 20:29:12.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:31:06.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.06s (1.90min) +2026-06-26 20:31:07.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:33:06.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:40:06.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 115, train_steps: 4600 +2026-06-26 20:40:06.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 20:40:07.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:40:07.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:41:12.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:41:12.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:43:00.484 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 20:43:01.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:44:55.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.71s (1.90min) +2026-06-26 20:44:56.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 20:46:53.373 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 20:53:52.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 116, train_steps: 4640 +2026-06-26 20:53:52.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 20:53:52.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:53:52.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:54:58.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 20:54:58.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 20:56:53.177 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 20:56:54.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 20:58:47.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.90s (1.88min) +2026-06-26 20:58:47.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:00:45.341 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:07:43.827 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 117, train_steps: 4680 +2026-06-26 21:07:44.519 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 21:07:44.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:07:44.737 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:08:49.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:08:49.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:10:41.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 21:10:42.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:12:34.885 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.15s (1.87min) +2026-06-26 21:12:35.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:14:32.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:21:30.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 118, train_steps: 4720 +2026-06-26 21:21:31.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 21:21:31.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:21:31.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:22:36.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:22:36.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:24:25.879 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 21:24:26.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:26:18.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.03s (1.87min) +2026-06-26 21:26:19.689 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:28:18.740 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:35:15.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 119, train_steps: 4760 +2026-06-26 21:35:16.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 21:35:16.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:35:17.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:36:22.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:36:22.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:38:10.121 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 21:38:11.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:40:04.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.48s (1.89min) +2026-06-26 21:40:05.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:42:05.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 21:49:05.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 120, train_steps: 4800 +2026-06-26 21:49:41.191 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=120 (missing=0, unexpected=0) +2026-06-26 21:49:43.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 21:49:43.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:49:43.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:50:50.867 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 21:50:50.868 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 21:52:37.690 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 21:52:38.807 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 21:54:30.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-26 21:54:31.611 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 21:56:28.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:03:25.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 121, train_steps: 4840 +2026-06-26 22:03:26.765 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 22:03:26.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:03:26.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:04:32.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:04:32.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:06:16.941 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 22:06:17.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:08:11.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-26 22:08:11.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:10:09.053 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:17:06.357 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 122, train_steps: 4880 +2026-06-26 22:17:07.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 22:17:07.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:17:07.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:18:12.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:18:12.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:19:58.738 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 22:19:59.709 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:21:52.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-26 22:21:53.536 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:23:50.578 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:30:48.493 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 123, train_steps: 4920 +2026-06-26 22:30:49.191 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 22:30:49.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:30:49.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:31:54.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:31:54.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:33:42.674 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 22:33:43.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:35:36.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.37s (1.87min) +2026-06-26 22:35:36.861 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:37:34.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:44:31.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 124, train_steps: 4960 +2026-06-26 22:44:32.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 22:44:32.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:44:32.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:45:37.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:45:37.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 22:47:23.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 22:47:24.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 22:49:16.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-26 22:49:17.480 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 22:51:15.770 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 22:58:14.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 125, train_steps: 5000 +2026-06-26 22:59:15.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 22:59:16.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 22:59:16.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:00:21.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:00:21.293 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:02:08.221 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 23:02:09.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:04:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-26 23:04:02.642 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:06:00.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:13:01.733 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 126, train_steps: 5040 +2026-06-26 23:13:02.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 23:13:02.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:13:02.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:14:07.946 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:14:07.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:15:52.138 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 23:15:53.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:17:45.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.46s (1.87min) +2026-06-26 23:17:46.393 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:19:44.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:26:43.670 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 127, train_steps: 5080 +2026-06-26 23:26:44.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 23:26:44.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:26:44.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:27:49.938 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:27:49.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:29:38.665 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 23:29:39.633 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:31:31.585 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.83s (1.86min) +2026-06-26 23:31:32.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:33:29.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:40:26.872 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 128, train_steps: 5120 +2026-06-26 23:40:27.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 23:40:27.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:40:27.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:41:32.908 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:41:32.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:43:25.296 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 23:43:26.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:45:18.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.57s (1.88min) +2026-06-26 23:45:19.671 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-26 23:47:18.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-26 23:54:15.224 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 129, train_steps: 5160 +2026-06-26 23:54:15.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-26 23:54:16.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:54:16.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:55:21.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-26 23:55:21.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-26 23:57:08.284 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-26 23:57:09.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-26 23:59:02.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.35s (1.87min) +2026-06-26 23:59:02.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:01:00.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:08:01.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 130, train_steps: 5200 +2026-06-27 00:08:37.934 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=130 (missing=0, unexpected=0) +2026-06-27 00:08:40.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 00:08:40.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:08:40.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:09:47.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:09:47.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:11:31.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 00:11:32.786 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:13:24.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 00:13:24.954 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:15:21.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:22:19.497 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 131, train_steps: 5240 +2026-06-27 00:22:20.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 00:22:20.501 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:22:20.596 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:23:25.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:23:25.537 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:25:14.491 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 00:25:15.497 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:27:08.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.18s (1.89min) +2026-06-27 00:27:09.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:29:06.969 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:36:05.698 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 132, train_steps: 5280 +2026-06-27 00:36:06.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 00:36:06.614 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:36:06.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:37:11.684 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:37:11.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:38:58.147 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 00:38:59.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:40:52.264 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 00:40:53.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:42:50.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 00:49:48.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 133, train_steps: 5320 +2026-06-27 00:49:49.281 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 00:49:49.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:49:49.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:50:54.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 00:50:54.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 00:52:43.945 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 00:52:44.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 00:54:36.504 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-27 00:54:37.231 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 00:56:35.586 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:03:34.399 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 134, train_steps: 5360 +2026-06-27 01:03:35.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 01:03:35.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:03:35.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:04:40.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:04:40.392 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:06:27.585 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 01:06:28.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:08:20.505 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 01:08:21.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:10:20.036 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:17:19.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 135, train_steps: 5400 +2026-06-27 01:17:19.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 01:17:19.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:17:20.015 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:18:25.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:18:25.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:20:14.470 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 01:20:15.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:22:09.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.56s (1.89min) +2026-06-27 01:22:09.844 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:24:07.321 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:31:06.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 136, train_steps: 5440 +2026-06-27 01:31:06.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 01:31:06.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:31:07.033 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:32:12.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:32:12.166 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:33:57.977 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 01:33:58.886 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:35:51.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.66s (1.88min) +2026-06-27 01:35:52.491 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:37:49.470 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:44:46.922 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 137, train_steps: 5480 +2026-06-27 01:44:47.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 01:44:47.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:44:47.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:45:53.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:45:53.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:47:39.516 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 01:47:40.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 01:49:31.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-27 01:49:32.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 01:51:29.596 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 01:58:26.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 138, train_steps: 5520 +2026-06-27 01:58:27.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 01:58:27.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:58:27.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 01:59:32.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 01:59:32.741 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:01:22.440 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 02:01:23.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:03:14.749 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.10s (1.85min) +2026-06-27 02:03:15.495 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:05:14.559 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:12:13.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 139, train_steps: 5560 +2026-06-27 02:12:14.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 02:12:14.852 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:12:14.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:13:20.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:13:20.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:15:08.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 02:15:09.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:17:02.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-27 02:17:02.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:19:00.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:26:01.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 140, train_steps: 5600 +2026-06-27 02:26:37.335 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=140 (missing=0, unexpected=0) +2026-06-27 02:26:39.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 02:26:39.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:26:39.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:27:46.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:27:46.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:29:32.904 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 02:29:33.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:31:26.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.05s (1.87min) +2026-06-27 02:31:26.957 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:33:23.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:40:22.100 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 141, train_steps: 5640 +2026-06-27 02:40:22.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 02:40:23.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:40:23.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:41:28.302 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:41:28.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:43:24.443 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 02:43:25.449 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:45:17.394 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.87s (1.86min) +2026-06-27 02:45:18.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 02:47:15.776 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 02:54:13.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 142, train_steps: 5680 +2026-06-27 02:54:14.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 02:54:14.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:54:14.779 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:55:19.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 02:55:19.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 02:57:08.127 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 02:57:09.303 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 02:59:01.043 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.55s (1.86min) +2026-06-27 02:59:01.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:01:00.447 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:08:00.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 143, train_steps: 5720 +2026-06-27 03:08:01.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 03:08:01.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:08:02.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:09:07.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:09:07.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:10:49.206 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 03:10:50.101 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:12:41.838 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 03:12:42.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:14:41.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:21:40.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 144, train_steps: 5760 +2026-06-27 03:21:41.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 03:21:41.688 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:21:41.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:22:47.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:22:47.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:24:37.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 03:24:38.522 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:26:30.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 03:26:30.737 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:28:27.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:35:25.728 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 145, train_steps: 5800 +2026-06-27 03:35:26.439 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 03:35:26.668 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:35:26.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:36:31.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:36:31.827 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:38:24.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 03:38:25.447 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:40:17.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-27 03:40:18.083 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:42:15.358 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 03:49:13.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 146, train_steps: 5840 +2026-06-27 03:49:14.441 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 03:49:14.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:49:14.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:50:19.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 03:50:19.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 03:52:07.384 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 03:52:08.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 03:54:00.384 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.67s (1.86min) +2026-06-27 03:54:01.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 03:56:00.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:03:02.506 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 147, train_steps: 5880 +2026-06-27 04:03:03.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 04:03:03.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:03:03.448 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:04:09.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:04:09.004 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:05:53.970 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 04:05:54.873 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:07:49.104 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.04s (1.90min) +2026-06-27 04:07:49.838 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:09:46.919 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:16:47.201 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 148, train_steps: 5920 +2026-06-27 04:16:47.920 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 04:16:48.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:16:48.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:17:53.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:17:53.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:19:41.577 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 04:19:42.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:21:34.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-27 04:21:34.914 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:23:32.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:30:30.111 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 149, train_steps: 5960 +2026-06-27 04:30:30.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 04:30:31.074 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:30:31.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:31:35.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:31:35.977 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:33:21.370 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 04:33:22.277 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:35:14.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.68s (1.86min) +2026-06-27 04:35:15.019 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:37:14.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:44:13.562 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 150, train_steps: 6000 +2026-06-27 04:45:56.511 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=150 (missing=0, unexpected=0) +2026-06-27 04:45:58.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 04:45:58.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:45:58.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:47:05.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:47:05.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 04:48:55.880 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 04:48:56.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 04:50:49.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.27s (1.87min) +2026-06-27 04:50:49.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 04:52:47.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 04:59:45.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 151, train_steps: 6040 +2026-06-27 04:59:46.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 04:59:46.753 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 04:59:46.788 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:00:52.086 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:00:52.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:02:39.060 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 05:02:40.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:04:31.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.00s (1.85min) +2026-06-27 05:04:31.951 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:06:29.400 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:13:26.875 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 152, train_steps: 6080 +2026-06-27 05:13:27.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 05:13:28.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:13:28.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:14:33.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:14:33.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:16:16.655 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 05:16:17.705 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:18:09.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 05:18:10.725 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:20:08.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:27:07.018 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 153, train_steps: 6120 +2026-06-27 05:27:07.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 05:27:07.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:27:07.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:28:13.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:28:13.063 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:30:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 05:30:01.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:31:53.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 05:31:54.526 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:33:51.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:40:49.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 154, train_steps: 6160 +2026-06-27 05:40:50.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 05:40:50.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:40:50.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:41:55.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:41:55.903 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:43:43.708 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 05:43:44.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:45:36.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.58s (1.86min) +2026-06-27 05:45:37.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 05:47:35.152 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 05:54:34.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 155, train_steps: 6200 +2026-06-27 05:54:35.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 05:54:35.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:54:35.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:55:40.254 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 05:55:40.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 05:57:29.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 05:57:30.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 05:59:21.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-27 05:59:22.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:01:20.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:08:18.608 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 156, train_steps: 6240 +2026-06-27 06:08:19.345 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 06:08:19.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:08:19.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:09:24.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:09:24.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:11:13.115 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 06:11:14.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:13:07.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.39s (1.89min) +2026-06-27 06:13:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:15:07.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:22:07.070 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 157, train_steps: 6280 +2026-06-27 06:22:07.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 06:22:07.998 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:22:08.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:23:13.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:23:13.083 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:24:58.737 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 06:24:59.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:26:53.287 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.45s (1.89min) +2026-06-27 06:26:54.017 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:28:51.195 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:35:49.190 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 158, train_steps: 6320 +2026-06-27 06:35:49.893 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 06:35:50.001 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:35:50.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:36:55.250 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:36:55.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:38:41.592 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 06:38:42.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:40:34.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.34s (1.86min) +2026-06-27 06:40:34.784 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:42:31.590 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 06:49:30.119 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 159, train_steps: 6360 +2026-06-27 06:49:30.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 06:49:30.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:49:31.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:50:35.926 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 06:50:35.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 06:52:20.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 06:52:21.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 06:54:13.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.82s (1.86min) +2026-06-27 06:54:14.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 06:56:13.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:03:11.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 160, train_steps: 6400 +2026-06-27 07:03:47.609 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=160 (missing=0, unexpected=0) +2026-06-27 07:03:49.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 07:03:49.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:03:49.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:04:56.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:04:56.141 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:06:39.597 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 07:06:40.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:08:32.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:08:33.334 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:10:30.349 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:17:28.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 161, train_steps: 6440 +2026-06-27 07:17:29.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 07:17:29.520 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:17:29.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:18:34.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:18:34.335 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:20:21.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 07:20:22.812 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:22:14.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-27 07:22:15.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:24:14.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:31:15.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 162, train_steps: 6480 +2026-06-27 07:31:16.120 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 07:31:16.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:31:16.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:32:21.511 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:32:21.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:34:11.154 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 07:34:12.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:36:05.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.52s (1.89min) +2026-06-27 07:36:06.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:38:06.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:45:06.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 163, train_steps: 6520 +2026-06-27 07:45:07.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 07:45:07.554 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:45:07.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:46:12.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:46:12.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:48:01.242 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 07:48:02.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 07:49:55.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 07:49:56.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 07:51:53.047 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 07:58:51.466 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 164, train_steps: 6560 +2026-06-27 07:58:52.198 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 07:58:52.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:58:52.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 07:59:57.267 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 07:59:57.269 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:01:45.187 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 08:01:46.382 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:03:38.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.75s (1.86min) +2026-06-27 08:03:39.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:05:37.929 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:12:37.141 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 165, train_steps: 6600 +2026-06-27 08:12:37.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 08:12:38.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:12:38.138 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:13:43.041 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:13:43.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:15:30.649 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 08:15:31.632 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:17:23.022 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 08:17:23.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:19:21.038 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:26:19.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 166, train_steps: 6640 +2026-06-27 08:26:20.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 08:26:20.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:26:20.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:27:25.850 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:27:25.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:29:09.907 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 08:29:10.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:31:03.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.96s (1.87min) +2026-06-27 08:31:03.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:33:00.862 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:40:02.145 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 167, train_steps: 6680 +2026-06-27 08:40:02.854 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 08:40:03.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:40:03.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:41:08.692 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:41:08.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:42:56.932 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 08:42:57.836 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:44:50.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.98s (1.88min) +2026-06-27 08:44:51.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 08:46:48.918 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 08:53:47.090 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 168, train_steps: 6720 +2026-06-27 08:53:47.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 08:53:47.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:53:47.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:54:53.142 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 08:54:53.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 08:56:49.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 08:56:50.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 08:58:42.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 08:58:42.853 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:00:42.006 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:07:40.863 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 169, train_steps: 6760 +2026-06-27 09:07:41.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 09:07:41.830 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:07:41.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:08:46.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:08:46.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:10:36.473 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 09:10:37.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:12:28.832 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 09:12:29.574 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:14:26.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:21:25.056 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 170, train_steps: 6800 +2026-06-27 09:22:00.629 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=170 (missing=0, unexpected=0) +2026-06-27 09:22:02.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 09:22:03.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:22:03.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:23:09.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:23:09.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:24:54.672 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 09:24:55.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:26:46.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.68s (1.84min) +2026-06-27 09:26:47.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:28:44.174 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:35:42.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 171, train_steps: 6840 +2026-06-27 09:35:43.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 09:35:43.318 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:35:43.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:36:48.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:36:48.483 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:38:34.283 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 09:38:35.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:40:27.278 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-27 09:40:28.046 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:42:25.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 09:49:25.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 172, train_steps: 6880 +2026-06-27 09:49:25.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 09:49:26.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:49:26.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:50:31.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 09:50:31.343 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 09:52:16.871 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 09:52:17.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 09:54:10.098 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.94s (1.87min) +2026-06-27 09:54:10.870 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 09:56:08.216 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:03:07.250 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 173, train_steps: 6920 +2026-06-27 10:03:07.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 10:03:08.080 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:03:08.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:04:12.968 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:04:12.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:05:59.863 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 10:06:00.818 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:07:54.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.05s (1.88min) +2026-06-27 10:07:54.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:09:52.462 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:16:52.270 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 174, train_steps: 6960 +2026-06-27 10:16:52.961 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 10:16:53.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:16:53.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:17:58.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:17:58.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:19:45.588 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 10:19:46.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:21:38.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.57s (1.86min) +2026-06-27 10:21:39.137 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:23:38.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:30:38.684 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 175, train_steps: 7000 +2026-06-27 10:31:49.736 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 10:31:49.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:31:50.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:32:55.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:32:55.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:34:42.767 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 10:34:43.678 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:36:35.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-27 10:36:36.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:38:33.381 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:45:31.037 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 176, train_steps: 7040 +2026-06-27 10:45:31.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 10:45:31.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:45:31.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:46:37.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:46:37.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 10:48:26.354 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 10:48:27.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 10:50:18.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 10:50:19.675 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 10:52:18.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 10:59:15.786 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 177, train_steps: 7080 +2026-06-27 10:59:16.464 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 10:59:16.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 10:59:16.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:00:22.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:00:22.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:02:12.070 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 11:02:13.081 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:04:06.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.00s (1.88min) +2026-06-27 11:04:06.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:06:06.390 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:13:06.245 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 178, train_steps: 7120 +2026-06-27 11:13:06.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 11:13:07.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:13:07.168 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:14:12.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:14:12.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:15:55.034 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 11:15:55.951 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:17:49.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.26s (1.89min) +2026-06-27 11:17:50.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:19:47.738 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:26:46.439 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 179, train_steps: 7160 +2026-06-27 11:26:47.157 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 11:26:47.369 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:26:47.477 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:27:52.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:27:52.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:29:40.799 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 11:29:41.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:31:33.613 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.40s (1.86min) +2026-06-27 11:31:34.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:33:31.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:40:28.518 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 180, train_steps: 7200 +2026-06-27 11:41:04.006 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=180 (missing=0, unexpected=0) +2026-06-27 11:41:06.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 11:41:06.183 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:41:06.237 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:42:12.139 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:42:12.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:43:58.131 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 11:43:59.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:45:52.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.10s (1.89min) +2026-06-27 11:45:53.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 11:47:50.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 11:54:47.136 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 181, train_steps: 7240 +2026-06-27 11:54:47.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 11:54:48.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:54:48.112 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:55:53.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 11:55:53.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 11:57:44.542 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 11:57:45.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 11:59:36.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-27 11:59:37.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:01:36.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:08:33.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 182, train_steps: 7280 +2026-06-27 12:08:33.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 12:08:34.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:08:34.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:09:39.270 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:09:39.274 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:11:26.700 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 12:11:27.717 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:13:19.515 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.59s (1.86min) +2026-06-27 12:13:20.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:15:19.835 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:22:18.651 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 183, train_steps: 7320 +2026-06-27 12:22:19.389 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 12:22:19.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:22:19.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:23:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:23:24.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:25:09.432 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 12:25:10.587 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:27:02.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 12:27:03.081 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:29:00.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:36:00.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 184, train_steps: 7360 +2026-06-27 12:36:01.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 12:36:01.401 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:36:01.476 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:37:06.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:37:06.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:38:52.462 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 12:38:53.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:40:44.844 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-27 12:40:45.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:42:44.004 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 12:49:43.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 185, train_steps: 7400 +2026-06-27 12:49:44.126 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 12:49:44.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:49:44.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:50:49.108 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 12:50:49.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 12:52:33.968 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 12:52:34.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 12:54:26.214 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.16s (1.85min) +2026-06-27 12:54:26.939 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 12:56:23.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:03:20.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 186, train_steps: 7440 +2026-06-27 13:03:21.296 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 13:03:21.435 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:03:21.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:04:26.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:04:26.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:06:12.581 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 13:06:13.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:08:06.422 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.75s (1.88min) +2026-06-27 13:08:07.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:10:06.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:17:06.958 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 187, train_steps: 7480 +2026-06-27 13:17:07.835 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 13:17:08.018 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:17:08.103 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:18:13.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:18:13.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:19:57.610 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 13:19:58.521 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:21:51.677 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.96s (1.88min) +2026-06-27 13:21:52.408 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:23:49.520 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:30:46.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 188, train_steps: 7520 +2026-06-27 13:30:47.055 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 13:30:47.258 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:30:47.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:31:52.164 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:31:52.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:33:46.093 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 13:33:47.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:35:38.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-27 13:35:39.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:37:38.823 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:44:37.406 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 189, train_steps: 7560 +2026-06-27 13:44:38.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 13:44:38.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:44:38.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:45:43.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:45:43.397 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 13:47:30.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 13:47:31.484 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 13:49:23.072 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 13:49:23.821 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 13:51:21.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 13:58:19.210 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 190, train_steps: 7600 +2026-06-27 13:58:55.252 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=190 (missing=0, unexpected=0) +2026-06-27 13:58:56.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 13:58:57.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 13:58:57.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:00:02.805 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:00:02.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:01:50.677 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 14:01:51.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:03:43.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.79s (1.86min) +2026-06-27 14:03:44.327 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:05:43.268 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:12:42.498 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 191, train_steps: 7640 +2026-06-27 14:12:43.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 14:12:43.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:12:43.404 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:13:48.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:13:48.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:15:31.388 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 14:15:32.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:17:24.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 14:17:24.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:19:22.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:26:19.397 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 192, train_steps: 7680 +2026-06-27 14:26:20.111 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 14:26:20.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:26:20.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:27:25.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:27:25.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:29:11.270 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 14:29:12.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:31:05.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.04s (1.88min) +2026-06-27 14:31:06.254 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:33:06.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:40:05.069 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 193, train_steps: 7720 +2026-06-27 14:40:05.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 14:40:05.966 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:40:06.069 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:41:11.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:41:11.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:42:57.220 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 14:42:58.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:44:51.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.89s (1.88min) +2026-06-27 14:44:51.963 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 14:46:48.869 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 14:53:46.414 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 194, train_steps: 7760 +2026-06-27 14:53:47.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 14:53:47.288 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:53:47.385 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:54:52.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 14:54:52.283 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 14:56:38.522 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 14:56:39.433 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 14:58:31.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-27 14:58:31.932 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:00:29.323 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:07:25.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 195, train_steps: 7800 +2026-06-27 15:07:26.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 15:07:26.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:07:26.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:08:32.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:08:32.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:10:17.063 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 15:10:18.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:12:10.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.25s (1.87min) +2026-06-27 15:12:11.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:14:08.837 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:21:05.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 196, train_steps: 7840 +2026-06-27 15:21:06.618 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 15:21:06.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:21:06.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:22:12.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:22:12.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:23:57.164 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 15:23:58.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:25:51.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.02s (1.88min) +2026-06-27 15:25:52.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:27:49.241 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:34:47.105 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 197, train_steps: 7880 +2026-06-27 15:34:47.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 15:34:47.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:34:48.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:35:53.245 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:35:53.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:37:43.288 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 15:37:44.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:39:36.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-27 15:39:37.123 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:41:34.407 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 15:48:31.332 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 198, train_steps: 7920 +2026-06-27 15:48:32.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 15:48:32.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:48:32.228 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:49:37.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 15:49:37.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 15:51:24.864 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 15:51:25.799 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 15:53:17.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-27 15:53:18.391 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 15:55:16.171 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:02:13.454 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 199, train_steps: 7960 +2026-06-27 16:02:14.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 16:02:14.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:02:14.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:03:19.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:03:19.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:05:04.157 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 16:05:05.089 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:06:56.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-27 16:06:57.376 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:08:54.464 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:15:51.547 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 200, train_steps: 8000 +2026-06-27 16:17:41.590 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=200 (missing=0, unexpected=0) +2026-06-27 16:17:43.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 16:17:43.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:17:43.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:18:50.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:18:50.036 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:20:36.935 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 16:20:37.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:22:29.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-27 16:22:30.101 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:24:27.952 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:31:25.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 201, train_steps: 8040 +2026-06-27 16:31:25.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 16:31:26.150 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:31:26.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:32:31.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:32:31.322 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:34:17.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 16:34:18.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:36:10.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-27 16:36:11.086 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:38:09.026 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:45:06.525 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 202, train_steps: 8080 +2026-06-27 16:45:07.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 16:45:07.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:45:07.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:46:12.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:46:12.701 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:47:58.208 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 16:47:59.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 16:49:52.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.88s (1.88min) +2026-06-27 16:49:52.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 16:51:49.983 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 16:58:46.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 203, train_steps: 8120 +2026-06-27 16:58:47.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 16:58:47.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:58:47.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 16:59:52.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 16:59:52.747 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:01:40.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 17:01:41.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:03:33.841 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-27 17:03:34.614 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:05:31.991 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:12:30.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 204, train_steps: 8160 +2026-06-27 17:12:31.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 17:12:31.294 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:12:31.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:13:36.635 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:13:36.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:15:21.401 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 17:15:22.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:17:14.686 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.08s (1.87min) +2026-06-27 17:17:15.432 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:19:14.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:26:14.267 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 205, train_steps: 8200 +2026-06-27 17:26:14.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 17:26:15.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:26:15.232 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:27:20.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:27:20.350 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:29:07.881 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 17:29:09.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:31:00.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 17:31:01.060 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:33:00.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:40:00.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 206, train_steps: 8240 +2026-06-27 17:40:00.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 17:40:00.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:40:01.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:41:06.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:41:06.091 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:42:54.369 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 17:42:55.359 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:44:49.417 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.95s (1.90min) +2026-06-27 17:44:50.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 17:46:47.075 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 17:53:44.816 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 207, train_steps: 8280 +2026-06-27 17:53:45.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 17:53:45.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:53:45.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:54:50.846 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 17:54:50.848 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 17:56:37.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 17:56:38.143 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 17:58:29.297 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.05s (1.85min) +2026-06-27 17:58:30.042 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:00:27.192 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:07:24.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 208, train_steps: 8320 +2026-06-27 18:07:25.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 18:07:25.630 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:07:25.715 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:08:30.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:08:30.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:10:22.824 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 18:10:24.076 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:12:15.543 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.33s (1.86min) +2026-06-27 18:12:16.291 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:14:15.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:21:13.791 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 209, train_steps: 8360 +2026-06-27 18:21:14.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 18:21:14.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:21:14.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:22:19.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:22:19.982 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:24:06.859 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 18:24:07.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:25:59.286 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.12s (1.85min) +2026-06-27 18:26:00.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:27:59.944 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:35:00.569 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 210, train_steps: 8400 +2026-06-27 18:35:36.756 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=210 (missing=0, unexpected=0) +2026-06-27 18:35:38.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 18:35:38.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:35:38.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:36:44.172 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:36:44.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:38:31.660 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 18:38:32.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:40:24.235 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-27 18:40:24.978 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:42:22.132 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 18:49:20.252 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 211, train_steps: 8440 +2026-06-27 18:49:21.035 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 18:49:21.203 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:49:21.257 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:50:26.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 18:50:26.376 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 18:52:15.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 18:52:16.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 18:54:08.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-27 18:54:09.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 18:56:07.094 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:03:05.723 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 212, train_steps: 8480 +2026-06-27 19:03:06.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 19:03:06.611 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:03:06.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:04:12.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:04:12.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:05:54.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 19:05:55.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:07:50.010 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.07s (1.90min) +2026-06-27 19:07:50.736 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:09:47.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:16:45.169 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 213, train_steps: 8520 +2026-06-27 19:16:45.871 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 19:16:46.037 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:16:46.106 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:17:51.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:17:51.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:19:41.804 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 19:19:42.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:21:34.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-27 19:21:34.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:23:32.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:30:29.667 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 214, train_steps: 8560 +2026-06-27 19:30:30.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 19:30:30.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:30:30.713 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:31:35.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:31:35.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:33:24.155 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 19:33:25.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:35:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-27 19:35:17.193 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:37:15.318 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:44:13.239 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 215, train_steps: 8600 +2026-06-27 19:44:13.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 19:44:14.163 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:44:14.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:45:19.194 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:45:19.195 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:47:08.993 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 19:47:10.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 19:49:02.211 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-27 19:49:02.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 19:51:00.896 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 19:58:01.579 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 216, train_steps: 8640 +2026-06-27 19:58:02.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 19:58:02.463 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:58:02.566 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 19:59:07.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 19:59:07.859 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:00:53.039 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 20:00:54.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:02:45.524 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.38s (1.86min) +2026-06-27 20:02:46.271 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:04:45.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:11:43.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 217, train_steps: 8680 +2026-06-27 20:11:43.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 20:11:43.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:11:43.962 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:12:49.304 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:12:49.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:14:38.101 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 20:14:39.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:16:30.473 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-27 20:16:31.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:18:28.666 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:25:25.309 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 218, train_steps: 8720 +2026-06-27 20:25:26.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 20:25:26.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:25:26.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:26:31.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:26:31.424 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:28:17.989 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 20:28:19.161 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:30:10.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.41s (1.86min) +2026-06-27 20:30:11.436 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:32:09.005 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:39:06.027 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 219, train_steps: 8760 +2026-06-27 20:39:06.730 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 20:39:06.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:39:07.042 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:40:12.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:40:12.244 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:42:03.411 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 20:42:04.531 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:43:55.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.91s (1.85min) +2026-06-27 20:43:56.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 20:45:53.751 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 20:52:51.442 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 220, train_steps: 8800 +2026-06-27 20:53:27.025 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=220 (missing=0, unexpected=0) +2026-06-27 20:53:29.196 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 20:53:29.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:53:29.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:54:34.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 20:54:34.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 20:56:20.307 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 20:56:21.292 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 20:58:13.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-27 20:58:13.734 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:00:12.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:07:09.778 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 221, train_steps: 8840 +2026-06-27 21:07:10.540 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 21:07:10.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:07:10.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:08:15.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:08:15.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:10:06.036 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 21:10:07.038 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:11:58.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-27 21:11:58.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:13:58.405 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:21:00.295 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 222, train_steps: 8880 +2026-06-27 21:21:01.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 21:21:01.122 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:21:01.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:22:06.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:22:06.275 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:23:51.010 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 21:23:52.012 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:25:43.249 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-27 21:25:43.973 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:27:42.598 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:34:41.459 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 223, train_steps: 8920 +2026-06-27 21:34:42.199 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 21:34:42.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:34:42.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:35:47.728 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:35:47.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:37:33.652 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 21:37:34.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:39:26.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 21:39:26.970 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:41:24.034 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 21:48:23.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 224, train_steps: 8960 +2026-06-27 21:48:23.843 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 21:48:24.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:48:24.053 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:49:29.619 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 21:49:29.622 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 21:51:14.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 21:51:15.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 21:53:08.100 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.04s (1.87min) +2026-06-27 21:53:08.864 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 21:55:07.207 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:02:06.463 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 225, train_steps: 9000 +2026-06-27 22:03:21.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 22:03:21.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:03:21.837 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:04:26.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:04:26.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:06:16.000 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 22:06:16.918 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:08:09.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.31s (1.87min) +2026-06-27 22:08:10.162 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:10:07.623 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:17:06.025 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 226, train_steps: 9040 +2026-06-27 22:17:06.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 22:17:06.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:17:07.094 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:18:11.782 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:18:11.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:19:56.153 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 22:19:57.075 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:21:49.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-27 22:21:50.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:23:47.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:30:46.830 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 227, train_steps: 9080 +2026-06-27 22:30:47.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 22:30:47.643 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:30:47.683 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:31:52.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:31:52.693 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:33:40.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 22:33:41.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:35:33.238 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-27 22:35:33.987 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:37:31.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:44:29.467 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 228, train_steps: 9120 +2026-06-27 22:44:30.207 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 22:44:30.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:44:30.365 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:45:35.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:45:35.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:47:17.271 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 22:47:18.344 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 22:49:09.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.11s (1.85min) +2026-06-27 22:49:10.346 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 22:51:07.488 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 22:58:06.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 229, train_steps: 9160 +2026-06-27 22:58:06.881 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 22:58:07.056 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:58:07.134 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 22:59:12.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 22:59:12.243 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:01:00.551 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 23:01:01.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:02:54.403 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-27 23:02:55.147 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:04:52.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:11:51.597 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 230, train_steps: 9200 +2026-06-27 23:12:27.536 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=230 (missing=0, unexpected=0) +2026-06-27 23:12:29.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 23:12:29.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:12:29.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:13:35.986 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:13:35.988 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:15:22.552 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 23:15:23.801 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:17:15.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-27 23:17:16.068 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:19:15.557 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:26:14.237 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 231, train_steps: 9240 +2026-06-27 23:26:15.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 23:26:15.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:26:15.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:27:20.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:27:20.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:29:05.917 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 23:29:06.840 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:30:58.364 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-27 23:30:59.117 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:32:59.223 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:40:01.412 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 232, train_steps: 9280 +2026-06-27 23:40:02.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 23:40:02.252 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:40:02.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:41:07.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:41:07.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:43:00.763 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 23:43:01.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:44:53.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.70s (1.86min) +2026-06-27 23:44:54.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-27 23:46:51.198 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-27 23:53:50.292 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 233, train_steps: 9320 +2026-06-27 23:53:51.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-27 23:53:51.154 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:53:51.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:54:56.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-27 23:54:56.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-27 23:56:43.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-27 23:56:44.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-27 23:58:36.046 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.43s (1.86min) +2026-06-27 23:58:36.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:00:34.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:07:31.985 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 234, train_steps: 9360 +2026-06-28 00:07:32.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 00:07:32.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:07:32.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:08:37.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:08:37.866 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:10:30.518 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 00:10:31.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:12:22.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 00:12:23.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:14:21.227 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:21:21.032 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 235, train_steps: 9400 +2026-06-28 00:21:21.762 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 00:21:21.913 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:21:21.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:22:27.224 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:22:27.227 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:24:11.389 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 00:24:12.306 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:26:05.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.27s (1.89min) +2026-06-28 00:26:06.505 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:28:05.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:35:06.581 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 236, train_steps: 9440 +2026-06-28 00:35:07.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 00:35:07.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:35:07.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:36:12.704 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:36:12.708 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:38:09.259 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 00:38:10.478 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:40:01.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 00:40:02.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:42:00.928 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 00:49:02.023 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 237, train_steps: 9480 +2026-06-28 00:49:02.723 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 00:49:02.937 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:49:03.047 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:50:08.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 00:50:08.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 00:51:56.103 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 00:51:57.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 00:53:50.146 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.84s (1.88min) +2026-06-28 00:53:50.905 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 00:55:48.013 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:02:47.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 238, train_steps: 9520 +2026-06-28 01:02:48.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 01:02:48.513 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:02:48.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:03:53.898 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:03:53.909 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:05:43.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 01:05:44.731 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:07:36.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.13s (1.85min) +2026-06-28 01:07:36.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:09:34.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:16:32.519 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 239, train_steps: 9560 +2026-06-28 01:16:33.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 01:16:33.459 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:16:33.560 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:17:38.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:17:38.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:19:30.419 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 01:19:31.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:21:22.589 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.07s (1.85min) +2026-06-28 01:21:23.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:23:21.074 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:30:20.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 240, train_steps: 9600 +2026-06-28 01:30:56.567 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=240 (missing=0, unexpected=0) +2026-06-28 01:30:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 01:30:58.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:30:58.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:32:04.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:32:04.363 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:33:51.671 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 01:33:52.609 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:35:44.059 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.27s (1.85min) +2026-06-28 01:35:44.824 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:37:43.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:44:43.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 241, train_steps: 9640 +2026-06-28 01:44:44.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 01:44:44.769 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:44:44.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:45:49.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:45:49.825 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:47:36.726 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 01:47:37.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 01:49:29.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.60s (1.86min) +2026-06-28 01:49:30.185 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 01:51:28.096 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 01:58:26.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 242, train_steps: 9680 +2026-06-28 01:58:27.402 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 01:58:27.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:58:27.653 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 01:59:32.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 01:59:32.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:01:17.694 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 02:01:18.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:03:10.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 02:03:11.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:05:08.163 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:12:07.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 243, train_steps: 9720 +2026-06-28 02:12:08.604 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 02:12:08.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:12:08.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:13:13.907 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:13:13.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:15:02.815 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 02:15:03.916 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:16:55.114 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.99s (1.85min) +2026-06-28 02:16:55.874 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:18:52.967 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:25:52.566 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 244, train_steps: 9760 +2026-06-28 02:25:53.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 02:25:53.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:25:53.538 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:26:58.739 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:26:58.740 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:28:44.615 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 02:28:45.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:30:37.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 02:30:37.801 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:32:37.556 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:39:37.858 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 245, train_steps: 9800 +2026-06-28 02:39:38.549 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 02:39:38.699 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:39:38.766 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:40:43.640 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:40:43.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:42:30.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 02:42:31.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:44:22.117 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.87s (1.85min) +2026-06-28 02:44:22.866 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 02:46:20.746 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 02:53:20.030 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 246, train_steps: 9840 +2026-06-28 02:53:20.735 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 02:53:20.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:53:21.019 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:54:26.309 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 02:54:26.311 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 02:56:12.404 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 02:56:13.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 02:58:06.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.09s (1.88min) +2026-06-28 02:58:07.355 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:00:06.612 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:07:07.313 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 247, train_steps: 9880 +2026-06-28 03:07:08.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 03:07:08.119 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:07:08.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:08:13.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:08:13.721 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:10:01.812 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 03:10:03.096 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:11:54.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:11:55.369 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:13:52.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:20:52.622 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 248, train_steps: 9920 +2026-06-28 03:20:53.331 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 03:20:53.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:20:53.525 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:21:58.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:21:58.797 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:23:47.607 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 03:23:48.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:25:41.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.88s (1.86min) +2026-06-28 03:25:41.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:27:39.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:34:39.417 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 249, train_steps: 9960 +2026-06-28 03:34:40.145 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 03:34:40.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:34:40.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:35:45.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:35:45.314 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:37:34.481 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 03:37:35.405 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:39:28.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.31s (1.89min) +2026-06-28 03:39:29.669 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:41:26.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 03:48:25.819 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 250, train_steps: 10000 +2026-06-28 03:50:11.901 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=250 (missing=0, unexpected=0) +2026-06-28 03:50:13.654 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 03:50:13.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:50:14.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:51:19.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 03:51:19.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 03:53:05.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 03:53:06.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 03:54:58.123 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.32s (1.86min) +2026-06-28 03:54:58.868 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 03:56:58.805 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:04:02.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 251, train_steps: 10040 +2026-06-28 04:04:02.896 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 04:04:02.994 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:04:03.044 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:05:08.673 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:05:08.675 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:06:55.564 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 04:06:56.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:08:50.295 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.62s (1.89min) +2026-06-28 04:08:51.040 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:10:48.404 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:17:47.916 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 252, train_steps: 10080 +2026-06-28 04:17:48.623 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 04:17:48.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:17:48.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:18:54.156 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:18:54.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:20:45.691 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 04:20:46.869 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:22:38.516 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-28 04:22:39.282 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:24:38.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:31:39.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 253, train_steps: 10120 +2026-06-28 04:31:40.230 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 04:31:40.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:31:40.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:32:45.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:32:45.562 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:34:28.604 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 04:34:29.517 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:36:21.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 04:36:21.907 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:38:20.382 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:45:19.594 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 254, train_steps: 10160 +2026-06-28 04:45:20.315 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 04:45:20.423 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:45:20.472 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:46:25.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:46:25.606 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 04:48:09.706 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 04:48:10.905 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 04:50:03.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.99s (1.87min) +2026-06-28 04:50:03.859 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 04:52:00.815 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 04:59:02.452 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 255, train_steps: 10200 +2026-06-28 04:59:03.182 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 04:59:03.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 04:59:03.488 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:00:08.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:00:08.746 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:01:51.517 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 05:01:52.426 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:03:44.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.42s (1.86min) +2026-06-28 05:03:44.771 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:05:43.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:12:43.964 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 256, train_steps: 10240 +2026-06-28 05:12:44.718 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 05:12:44.930 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:12:45.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:13:49.999 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:13:50.003 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:15:34.016 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 05:15:34.935 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:17:26.657 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.49s (1.86min) +2026-06-28 05:17:27.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:19:24.962 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:26:25.194 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 257, train_steps: 10280 +2026-06-28 05:26:25.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 05:26:26.110 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:26:26.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:27:31.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:27:31.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:29:21.044 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 05:29:21.967 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:31:13.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 05:31:14.367 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:33:13.735 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:40:14.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 258, train_steps: 10320 +2026-06-28 05:40:14.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 05:40:15.148 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:40:15.279 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:41:20.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:41:20.418 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:43:09.829 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 05:43:10.752 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:45:05.021 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 114.10s (1.90min) +2026-06-28 05:45:05.787 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 05:47:05.628 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 05:54:06.128 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 259, train_steps: 10360 +2026-06-28 05:54:06.824 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 05:54:06.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:54:06.983 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:55:12.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 05:55:12.213 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 05:56:59.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 05:57:00.470 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 05:58:52.890 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.24s (1.87min) +2026-06-28 05:58:53.634 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:00:50.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:07:48.832 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 260, train_steps: 10400 +2026-06-28 06:08:24.813 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=260 (missing=0, unexpected=0) +2026-06-28 06:08:27.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 06:08:27.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:08:27.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:09:32.720 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:09:32.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:11:20.207 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 06:11:21.247 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:13:12.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 06:13:13.629 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:15:11.532 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:22:10.900 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 261, train_steps: 10440 +2026-06-28 06:22:11.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 06:22:11.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:22:11.833 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:23:16.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:23:16.980 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:25:06.072 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 06:25:06.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:26:58.371 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 06:26:59.134 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:28:58.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:36:01.351 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 262, train_steps: 10480 +2026-06-28 06:36:02.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 06:36:02.192 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:36:02.248 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:37:08.432 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:37:08.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:38:52.554 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 06:38:53.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:40:45.118 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 06:40:45.857 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:42:43.996 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 06:49:44.847 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 263, train_steps: 10520 +2026-06-28 06:49:45.559 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 06:49:45.760 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:49:45.856 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:50:50.751 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 06:50:50.757 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 06:52:36.808 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 06:52:37.732 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 06:54:29.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 06:54:30.155 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 06:56:27.179 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:03:27.003 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 264, train_steps: 10560 +2026-06-28 07:03:27.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 07:03:27.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:03:27.811 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:04:32.919 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:04:32.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:06:20.617 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 07:06:21.536 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:08:13.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.46s (1.86min) +2026-06-28 07:08:13.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:10:12.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:17:11.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 265, train_steps: 10600 +2026-06-28 07:17:12.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 07:17:12.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:17:12.319 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:18:17.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:18:17.193 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:20:03.095 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 07:20:04.204 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:21:55.666 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.28s (1.85min) +2026-06-28 07:21:56.435 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:23:53.613 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:30:52.977 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 266, train_steps: 10640 +2026-06-28 07:30:53.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 07:30:53.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:30:53.970 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:31:58.939 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:31:58.940 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:33:47.057 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 07:33:48.071 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:35:39.487 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 07:35:40.255 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:37:38.702 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:44:39.465 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 267, train_steps: 10680 +2026-06-28 07:44:40.187 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 07:44:40.340 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:44:40.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:45:45.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:45:45.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:47:32.045 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 07:47:32.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 07:49:24.593 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 07:49:25.339 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 07:51:22.935 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 07:58:20.921 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 268, train_steps: 10720 +2026-06-28 07:58:21.655 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 07:58:21.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:58:21.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 07:59:27.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 07:59:27.034 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:01:15.619 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 08:01:16.546 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:03:09.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.73s (1.88min) +2026-06-28 08:03:10.259 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:05:07.461 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:12:07.911 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 269, train_steps: 10760 +2026-06-28 08:12:08.563 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 08:12:08.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:12:08.862 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:13:14.027 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:13:14.030 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:15:02.253 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 08:15:03.425 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:16:55.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.53s (1.86min) +2026-06-28 08:16:55.854 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:18:53.473 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:25:53.524 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 270, train_steps: 10800 +2026-06-28 08:26:29.738 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=270 (missing=0, unexpected=0) +2026-06-28 08:26:31.373 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 08:26:31.547 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:26:31.637 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:27:36.959 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:27:36.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:29:26.967 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 08:29:27.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:31:19.469 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.39s (1.86min) +2026-06-28 08:31:20.233 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:33:19.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:40:21.902 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 271, train_steps: 10840 +2026-06-28 08:40:22.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 08:40:22.865 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:40:22.902 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:41:28.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:41:28.239 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:43:13.075 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 08:43:14.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:45:07.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.43s (1.89min) +2026-06-28 08:45:08.333 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 08:47:06.762 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 08:54:07.116 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 272, train_steps: 10880 +2026-06-28 08:54:07.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 08:54:07.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:54:08.222 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:55:13.396 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 08:55:13.400 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 08:56:59.167 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 08:57:00.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 08:58:52.691 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.42s (1.87min) +2026-06-28 08:58:53.455 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:00:50.701 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:07:50.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 273, train_steps: 10920 +2026-06-28 09:07:50.777 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 09:07:50.960 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:07:51.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:08:56.054 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:08:56.058 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:10:42.956 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 09:10:43.878 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:12:35.629 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 09:12:36.380 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:14:33.677 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:21:31.842 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 274, train_steps: 10960 +2026-06-28 09:21:32.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 09:21:32.667 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:21:32.696 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:22:38.105 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:22:38.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:24:27.534 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 09:24:28.458 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:26:20.474 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.84s (1.86min) +2026-06-28 09:26:21.228 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:28:20.028 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:35:19.709 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 275, train_steps: 11000 +2026-06-28 09:36:34.979 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 09:36:35.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:36:35.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:37:40.407 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:37:40.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:39:26.711 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 09:39:27.644 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:41:19.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 09:41:20.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:43:19.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 09:50:21.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 276, train_steps: 11040 +2026-06-28 09:50:22.434 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 09:50:22.550 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:50:22.591 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:51:27.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 09:51:27.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 09:53:23.988 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 09:53:25.064 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 09:55:17.109 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.95s (1.87min) +2026-06-28 09:55:17.850 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 09:57:16.164 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:04:15.620 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 277, train_steps: 11080 +2026-06-28 10:04:16.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 10:04:16.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:04:16.485 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:05:21.745 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:05:21.750 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:07:10.872 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 10:07:11.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:09:04.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.97s (1.88min) +2026-06-28 10:09:05.713 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:11:05.540 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:18:08.486 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 278, train_steps: 11120 +2026-06-28 10:18:09.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 10:18:09.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:18:09.620 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:19:14.767 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:19:14.771 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:21:00.797 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 10:21:02.241 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:22:54.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.19s (1.87min) +2026-06-28 10:22:55.398 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:24:52.743 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:31:53.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 279, train_steps: 11160 +2026-06-28 10:31:54.009 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 10:31:54.121 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:31:54.167 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:32:59.131 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:32:59.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:34:53.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 10:34:53.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:36:45.569 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.45s (1.86min) +2026-06-28 10:36:46.314 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:38:43.860 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 10:45:42.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 280, train_steps: 11200 +2026-06-28 10:46:18.826 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=280 (missing=0, unexpected=0) +2026-06-28 10:46:20.610 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 10:46:20.734 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:46:20.784 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:47:26.375 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 10:47:26.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 10:49:12.349 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 10:49:13.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 10:51:06.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.08s (1.88min) +2026-06-28 10:51:07.307 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 10:53:07.146 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:00:07.439 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 281, train_steps: 11240 +2026-06-28 11:00:08.169 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 11:00:08.299 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:00:08.360 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:01:13.320 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:01:13.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:02:59.509 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 11:03:00.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:04:53.445 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.62s (1.88min) +2026-06-28 11:04:54.184 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:06:51.371 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:13:50.645 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 282, train_steps: 11280 +2026-06-28 11:13:51.391 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 11:13:51.588 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:13:51.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:14:56.813 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:14:56.814 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:16:42.603 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 11:16:43.526 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:18:35.265 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 11:18:36.016 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:20:33.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:27:31.834 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 283, train_steps: 11320 +2026-06-28 11:27:32.532 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 11:27:32.716 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:27:32.780 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:28:37.973 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:28:37.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:30:22.262 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 11:30:23.414 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:32:14.955 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 11:32:15.708 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:34:14.695 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:41:14.910 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 284, train_steps: 11360 +2026-06-28 11:41:15.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 11:41:15.792 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:41:15.891 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:42:20.822 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:42:20.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:44:07.566 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 11:44:08.706 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:46:01.206 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.13s (1.87min) +2026-06-28 11:46:01.966 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 11:48:00.476 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 11:55:02.699 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 285, train_steps: 11400 +2026-06-28 11:55:03.421 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 11:55:03.594 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:55:03.687 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:56:08.917 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 11:56:08.921 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 11:57:56.427 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 11:57:57.455 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 11:59:50.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.81s (1.88min) +2026-06-28 11:59:51.106 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:01:48.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:08:47.897 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 286, train_steps: 11440 +2026-06-28 12:08:48.608 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 12:08:48.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:08:48.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:09:54.066 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:09:54.067 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:11:46.963 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 12:11:48.087 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:13:39.506 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.29s (1.85min) +2026-06-28 12:13:40.269 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:15:39.203 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:22:39.790 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 287, train_steps: 11480 +2026-06-28 12:22:40.529 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 12:22:40.638 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:22:40.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:23:45.847 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:23:45.851 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:25:33.885 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 12:25:34.874 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:27:27.325 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.33s (1.87min) +2026-06-28 12:27:28.071 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:29:25.085 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:36:23.659 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 288, train_steps: 11520 +2026-06-28 12:36:24.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 12:36:24.561 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:36:24.642 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:37:29.652 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:37:29.659 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:39:16.353 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 12:39:17.324 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:41:09.681 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.18s (1.87min) +2026-06-28 12:41:10.449 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:43:07.885 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 12:50:07.855 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 289, train_steps: 11560 +2026-06-28 12:50:08.712 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 12:50:08.826 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:50:08.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:51:14.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 12:51:14.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 12:53:03.900 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 12:53:04.901 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 12:54:57.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.00s (1.87min) +2026-06-28 12:54:57.753 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 12:56:55.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:03:54.704 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 290, train_steps: 11600 +2026-06-28 13:04:30.693 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=290 (missing=0, unexpected=0) +2026-06-28 13:04:33.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 13:04:33.273 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:04:33.307 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:05:39.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:05:39.229 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:07:33.019 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 13:07:33.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:09:25.742 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.61s (1.86min) +2026-06-28 13:09:26.503 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:11:23.644 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:18:21.781 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 291, train_steps: 11640 +2026-06-28 13:18:22.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 13:18:22.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:18:22.756 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:19:28.153 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:19:28.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:21:13.299 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 13:21:14.308 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:23:07.648 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.24s (1.89min) +2026-06-28 13:23:08.411 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:25:07.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:32:06.325 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 292, train_steps: 11680 +2026-06-28 13:32:07.130 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 13:32:07.233 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:32:07.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:33:12.411 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:33:12.412 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:34:59.612 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 13:35:00.855 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:36:53.282 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.41s (1.87min) +2026-06-28 13:36:54.035 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:38:51.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:45:49.093 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 293, train_steps: 11720 +2026-06-28 13:45:49.796 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 13:45:49.949 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:45:50.013 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:46:55.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:46:55.225 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 13:48:41.545 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 13:48:42.628 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 13:50:34.481 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 13:50:35.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 13:52:32.277 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 13:59:30.024 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 294, train_steps: 11760 +2026-06-28 13:59:30.793 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 13:59:30.991 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 13:59:31.099 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:00:36.174 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:00:36.177 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:02:20.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 14:02:21.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:04:12.509 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.17s (1.85min) +2026-06-28 14:04:13.258 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:06:10.182 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:13:07.553 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 295, train_steps: 11800 +2026-06-28 14:13:08.291 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 14:13:08.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:13:08.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:14:13.535 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:14:13.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:16:08.492 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 14:16:09.685 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:18:00.558 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.74s (1.85min) +2026-06-28 14:18:01.316 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:20:00.633 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:27:01.394 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 296, train_steps: 11840 +2026-06-28 14:27:02.095 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 14:27:02.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:27:02.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:28:07.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:28:07.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:29:50.657 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 14:29:51.645 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:31:42.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.20s (1.85min) +2026-06-28 14:31:43.703 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:33:42.499 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:40:43.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 297, train_steps: 11880 +2026-06-28 14:40:43.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 14:40:43.978 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:40:44.031 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:41:49.073 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:41:49.079 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:43:34.976 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 14:43:35.987 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:45:27.346 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.26s (1.85min) +2026-06-28 14:45:28.113 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 14:47:25.489 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 14:54:24.523 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 298, train_steps: 11920 +2026-06-28 14:54:25.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 14:54:25.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:54:25.533 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:55:30.626 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 14:55:30.627 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 14:57:14.819 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 14:57:15.828 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 14:59:08.527 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.60s (1.88min) +2026-06-28 14:59:09.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:01:07.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:08:06.469 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 299, train_steps: 11960 +2026-06-28 15:08:07.178 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 15:08:07.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:08:07.431 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:09:12.377 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:09:12.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:11:01.008 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 15:11:02.255 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:12:53.823 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.56s (1.86min) +2026-06-28 15:12:54.582 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:14:51.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:21:49.812 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 300, train_steps: 12000 +2026-06-28 15:23:34.019 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=300 (missing=0, unexpected=0) +2026-06-28 15:23:36.049 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 15:23:36.173 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:23:36.223 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:24:41.310 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:24:41.313 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:26:31.035 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 15:26:31.770 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:28:23.465 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 15:28:24.215 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:30:21.384 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:37:20.383 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 301, train_steps: 12040 +2026-06-28 15:37:21.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 15:37:21.316 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:37:21.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:38:26.093 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:38:26.097 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:40:13.962 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 15:40:14.900 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:42:06.710 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:42:07.451 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:44:06.512 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 15:51:07.501 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 302, train_steps: 12080 +2026-06-28 15:51:08.210 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 15:51:08.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:51:08.416 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:52:13.820 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 15:52:13.821 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 15:54:03.550 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 15:54:04.861 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 15:55:56.707 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.64s (1.86min) +2026-06-28 15:55:57.450 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 15:57:55.177 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:04:53.635 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 303, train_steps: 12120 +2026-06-28 16:04:54.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 16:04:54.499 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:04:54.576 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:05:59.508 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:05:59.512 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:07:49.263 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 16:07:50.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:09:41.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 16:09:42.587 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:11:41.072 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:18:40.062 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 304, train_steps: 12160 +2026-06-28 16:18:40.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 16:18:40.943 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:18:40.985 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:19:46.215 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:19:46.219 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:21:33.374 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 16:21:34.388 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:23:25.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.22s (1.85min) +2026-06-28 16:23:26.478 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:25:23.602 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:32:22.212 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 305, train_steps: 12200 +2026-06-28 16:32:22.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 16:32:23.135 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:32:23.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:33:28.486 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:33:28.490 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:35:13.430 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 16:35:14.427 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:37:06.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.89s (1.86min) +2026-06-28 16:37:07.200 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:39:06.217 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:46:06.534 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 306, train_steps: 12240 +2026-06-28 16:46:07.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 16:46:07.415 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:46:07.492 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:47:12.722 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:47:12.726 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 16:48:59.323 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 16:49:00.263 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 16:50:52.953 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.48s (1.87min) +2026-06-28 16:50:53.717 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 16:52:50.889 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 16:59:49.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 307, train_steps: 12280 +2026-06-28 16:59:50.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 16:59:50.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 16:59:50.468 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:00:55.383 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:00:55.395 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:02:43.666 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 17:02:44.612 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:04:36.285 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 17:04:37.054 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:06:34.912 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:13:33.682 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 308, train_steps: 12320 +2026-06-28 17:13:34.413 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 17:13:34.545 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:13:34.590 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:14:39.679 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:14:39.680 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:16:30.954 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 17:16:31.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:18:22.996 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.92s (1.85min) +2026-06-28 17:18:23.745 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:20:20.828 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:27:20.300 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 309, train_steps: 12360 +2026-06-28 17:27:21.006 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 17:27:21.181 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:27:21.260 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:28:26.329 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:28:26.333 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:30:17.630 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 17:30:18.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:32:10.597 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.54s (1.86min) +2026-06-28 17:32:11.365 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:34:08.515 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:41:06.718 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 310, train_steps: 12400 +2026-06-28 17:41:42.650 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=310 (missing=0, unexpected=0) +2026-06-28 17:41:44.575 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 17:41:44.724 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:41:44.783 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:42:50.029 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:42:50.032 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:44:42.415 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 17:44:43.337 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 17:46:34.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.04s (1.85min) +2026-06-28 17:46:35.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 17:48:32.545 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 17:55:30.917 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 311, train_steps: 12440 +2026-06-28 17:55:31.725 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 17:55:31.834 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:55:31.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:56:36.911 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 17:56:36.914 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 17:58:24.596 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 17:58:25.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:00:17.025 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.31s (1.86min) +2026-06-28 18:00:17.795 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:02:16.243 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:09:14.915 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 312, train_steps: 12480 +2026-06-28 18:09:15.778 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 18:09:15.879 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:09:15.922 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:10:21.268 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:10:21.272 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:12:09.158 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 18:12:10.460 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:14:02.437 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.80s (1.86min) +2026-06-28 18:14:03.206 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:16:01.082 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:23:02.370 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 313, train_steps: 12520 +2026-06-28 18:23:03.061 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 18:23:03.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:23:03.217 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:24:08.759 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:24:08.763 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:25:53.373 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 18:25:54.349 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:27:45.972 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.48s (1.86min) +2026-06-28 18:27:46.742 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:29:44.764 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:36:44.097 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 314, train_steps: 12560 +2026-06-28 18:36:44.785 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 18:36:44.882 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:36:44.929 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:37:50.132 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:37:50.136 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:39:38.231 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 18:39:39.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:41:30.361 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.02s (1.85min) +2026-06-28 18:41:31.126 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:43:28.236 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 18:50:25.530 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 315, train_steps: 12600 +2026-06-28 18:50:26.240 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 18:50:26.406 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:50:26.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:51:31.475 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 18:51:31.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 18:53:21.319 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 18:53:22.234 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 18:55:13.829 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.23s (1.85min) +2026-06-28 18:55:14.583 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 18:57:13.934 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:04:13.078 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 316, train_steps: 12640 +2026-06-28 19:04:13.776 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 19:04:13.974 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:04:14.070 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:05:19.351 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:05:19.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:07:07.110 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 19:07:08.051 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:08:59.910 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.52s (1.86min) +2026-06-28 19:09:00.681 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:11:00.539 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:18:01.149 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 317, train_steps: 12680 +2026-06-28 19:18:01.876 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 19:18:02.068 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:18:02.175 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:19:07.202 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:19:07.205 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:20:52.664 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 19:20:53.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:22:44.733 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 19:22:45.492 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:24:44.249 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:31:43.591 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 318, train_steps: 12720 +2026-06-28 19:31:44.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 19:31:44.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:31:44.664 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:32:49.802 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:32:49.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:34:36.663 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 19:34:37.670 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:36:28.367 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.59s (1.84min) +2026-06-28 19:36:29.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:38:26.150 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:45:24.088 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 319, train_steps: 12760 +2026-06-28 19:45:24.787 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 19:45:24.944 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:45:25.008 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:46:30.016 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:46:30.020 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 19:48:17.816 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 19:48:18.884 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 19:50:10.500 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.47s (1.86min) +2026-06-28 19:50:11.263 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 19:52:08.431 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 19:59:06.558 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 320, train_steps: 12800 +2026-06-28 19:59:42.298 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=320 (missing=0, unexpected=0) +2026-06-28 19:59:44.368 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 19:59:44.494 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 19:59:44.557 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:00:49.927 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:00:49.934 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:02:34.356 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 20:02:35.357 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:04:26.410 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:04:27.156 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:06:24.865 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:13:24.095 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 321, train_steps: 12840 +2026-06-28 20:13:24.806 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 20:13:24.928 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:13:24.976 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:14:29.993 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:14:29.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:16:15.047 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 20:16:16.057 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:18:07.672 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.50s (1.86min) +2026-06-28 20:18:08.403 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:20:06.646 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:27:06.409 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 322, train_steps: 12880 +2026-06-28 20:27:07.128 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 20:27:07.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:27:07.300 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:28:12.140 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:28:12.144 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:29:58.446 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 20:29:59.453 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:31:51.995 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.44s (1.87min) +2026-06-28 20:31:52.754 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:33:50.099 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:40:48.242 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 323, train_steps: 12920 +2026-06-28 20:40:48.957 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 20:40:49.052 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:40:49.116 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:41:54.370 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:41:54.374 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:43:45.328 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 20:43:46.436 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:45:37.482 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.95s (1.85min) +2026-06-28 20:45:38.256 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 20:47:37.204 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 20:54:35.785 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 324, train_steps: 12960 +2026-06-28 20:54:36.493 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 20:54:36.607 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:54:36.651 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:55:41.574 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 20:55:41.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 20:57:33.434 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 20:57:34.430 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 20:59:25.804 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.24s (1.85min) +2026-06-28 20:59:26.573 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:01:23.649 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:08:23.487 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 325, train_steps: 13000 +2026-06-28 21:09:41.409 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 21:09:41.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:09:41.647 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:10:47.162 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:10:47.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:12:34.644 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 21:12:35.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:14:26.624 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.85s (1.85min) +2026-06-28 21:14:27.396 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:16:24.724 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:23:23.153 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 326, train_steps: 13040 +2026-06-28 21:23:23.853 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 21:23:23.950 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:23:24.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:24:29.184 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:24:29.190 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:26:14.846 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 21:26:15.875 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:28:07.658 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.69s (1.86min) +2026-06-28 21:28:08.426 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:30:06.650 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:37:06.714 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 327, train_steps: 13080 +2026-06-28 21:37:07.399 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 21:37:07.577 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:37:07.656 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:38:12.714 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:38:12.719 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:39:57.613 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 21:39:58.564 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:41:51.358 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.61s (1.88min) +2026-06-28 21:41:52.108 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:43:49.768 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 21:50:48.002 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 328, train_steps: 13120 +2026-06-28 21:50:48.690 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 21:50:48.872 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:50:48.969 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:51:54.209 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 21:51:54.212 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 21:53:42.340 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 21:53:43.390 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 21:55:34.674 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 21:55:35.428 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 21:57:33.337 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:04:31.084 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 329, train_steps: 13160 +2026-06-28 22:04:31.789 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 22:04:31.945 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:04:32.000 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:05:37.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:05:37.007 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:07:25.750 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 22:07:26.698 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:09:18.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.36s (1.86min) +2026-06-28 22:09:18.979 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:11:17.521 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:18:14.766 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 330, train_steps: 13200 +2026-06-28 22:18:50.802 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=330 (missing=0, unexpected=0) +2026-06-28 22:18:52.997 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 22:18:53.129 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:18:53.170 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:19:58.565 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:19:58.568 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:21:46.145 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 22:21:47.253 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:23:39.159 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.73s (1.86min) +2026-06-28 22:23:40.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:25:39.041 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:32:39.441 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 331, train_steps: 13240 +2026-06-28 22:32:40.185 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 22:32:40.336 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:32:40.378 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:33:45.600 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:33:45.603 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:35:34.285 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 22:35:35.226 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:37:27.127 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.72s (1.86min) +2026-06-28 22:37:27.891 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:39:25.154 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 22:46:23.348 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 332, train_steps: 13280 +2026-06-28 22:46:24.045 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 22:46:24.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:46:24.342 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:47:29.155 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 22:47:29.160 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 22:49:17.479 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 22:49:18.580 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 22:51:10.541 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.74s (1.86min) +2026-06-28 22:51:11.310 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 22:53:08.446 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:00:06.352 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 333, train_steps: 13320 +2026-06-28 23:00:07.002 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 23:00:07.236 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:00:07.347 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:01:12.197 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:01:12.201 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:02:58.269 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 23:02:59.186 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:04:53.028 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 113.65s (1.89min) +2026-06-28 23:04:53.783 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:06:50.925 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:13:49.022 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 334, train_steps: 13360 +2026-06-28 23:13:49.743 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 23:13:49.904 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:13:49.963 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:14:54.956 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:14:54.965 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:16:43.420 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 23:16:44.420 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:18:35.880 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.35s (1.86min) +2026-06-28 23:18:36.665 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:20:33.533 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:27:31.173 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 335, train_steps: 13400 +2026-06-28 23:27:31.990 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 23:27:32.165 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:27:32.246 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:28:37.467 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:28:37.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:30:22.092 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 23:30:23.124 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:32:14.539 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.19s (1.85min) +2026-06-28 23:32:15.279 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:34:14.643 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:41:13.343 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 336, train_steps: 13440 +2026-06-28 23:41:14.039 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 23:41:14.137 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:41:14.179 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:42:19.503 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:42:19.507 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:44:05.565 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 23:44:06.491 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:45:57.819 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.14s (1.85min) +2026-06-28 23:45:58.603 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-28 23:47:57.887 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-28 23:54:59.360 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 337, train_steps: 13480 +2026-06-28 23:55:00.090 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-28 23:55:00.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:55:00.266 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:56:05.276 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-28 23:56:05.280 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-28 23:57:48.918 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-28 23:57:49.924 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-28 23:59:41.040 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.01s (1.85min) +2026-06-28 23:59:41.804 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:01:39.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:08:38.647 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 338, train_steps: 13520 +2026-06-29 00:08:39.356 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-29 00:08:39.480 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:08:39.528 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:09:44.548 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:09:44.552 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:11:28.122 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-29 00:11:29.151 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:13:20.176 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.89s (1.85min) +2026-06-29 00:13:20.937 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:15:19.350 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:22:18.604 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 339, train_steps: 13560 +2026-06-29 00:22:19.332 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-29 00:22:19.451 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:22:19.495 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:23:24.774 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:23:24.775 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:25:12.501 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-29 00:25:13.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:27:05.328 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.71s (1.86min) +2026-06-29 00:27:06.087 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:29:04.997 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:36:05.214 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 340, train_steps: 13600 +2026-06-29 00:36:41.381 | INFO | __main__:maybe_update_ref_model_for_kl:804 - [Moving KL] Updated ref_model at update_step=340 (missing=0, unexpected=0) +2026-06-29 00:36:43.011 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-29 00:36:43.115 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:36:43.188 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:37:48.438 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:37:48.442 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:39:38.560 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-29 00:39:39.553 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:41:30.479 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 110.79s (1.85min) +2026-06-29 00:41:31.232 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:43:28.471 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 00:50:26.415 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 341, train_steps: 13640 +2026-06-29 00:50:27.221 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-29 00:50:27.338 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:50:27.380 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:51:32.694 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 00:51:32.697 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 00:53:18.757 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-29 00:53:19.711 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 00:55:12.062 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 112.16s (1.87min) +2026-06-29 00:55:12.848 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-06-29 00:57:10.329 | INFO | hyvideo.utils.grpo_utils:train_one_step:2941 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +2026-06-29 01:04:08.474 | INFO | hyvideo.utils.grpo_utils:train_one_step:3585 - Rank 9, sp_rank 0: update step: 342, train_steps: 13680 +2026-06-29 01:04:09.189 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1261 - [ParallelGroups] Rank 9, dp_rank 4: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=2, rank_in_group=0 +2026-06-29 01:04:09.381 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:04:09.498 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:05:14.598 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1399 - Rank 9 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-06-29 01:05:14.602 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1424 - Rank 9 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +2026-06-29 01:07:02.832 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:1033 - [GatherRewards] Rank 9: Using local metric names without global sync: ['avg'] +2026-06-29 01:07:03.971 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1819 - Rank 9: Pre-computing Reference Model statistics (Batch-wise)... +2026-06-29 01:08:55.572 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1962 - Rank 9: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 111.44s (1.86min) +2026-06-29 01:08:56.335 | INFO | hyvideo.utils.grpo_utils:train_one_step:2713 - Rank 9, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, spatial_granularity=frame, advantage_source=token, shuffle_weights=False, uniform_weights=True. diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/debug.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/debug.log new file mode 100644 index 0000000000000000000000000000000000000000..e003af185b81532cacc394df57802192ccb0c74f --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/debug.log @@ -0,0 +1,19 @@ +2026-06-25 18:01:16,347 INFO MainThread:1812 [wandb_setup.py:_flush():81] Current SDK version is 0.28.0 +2026-06-25 18:01:16,347 INFO MainThread:1812 [wandb_setup.py:_flush():81] Configure stats pid to 1812 +2026-06-25 18:01:16,348 INFO MainThread:1812 [wandb_setup.py:_flush():81] Loading settings from environment variables +2026-06-25 18:01:16,348 INFO MainThread:1812 [wandb_init.py:setup_run_log_directory():725] Logging user logs to outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/run-20260625_180116-dgy7zq92/logs/debug.log +2026-06-25 18:01:16,348 INFO MainThread:1812 [wandb_init.py:setup_run_log_directory():726] Logging internal logs to outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/run-20260625_180116-dgy7zq92/logs/debug-internal.log +2026-06-25 18:01:16,348 INFO MainThread:1812 [wandb_init.py:init():768] calling init triggers +2026-06-25 18:01:16,348 INFO MainThread:1812 [wandb_init.py:init():773] wandb.init called with sweep_config: {} +config: {'pretrained_model_root': './ckpts', 'pretrained_transformer_version': '480p_t2v', 'post_train_type': 'grpo', 'learning_rate': 1e-05, 'weight_decay': 0.01, 'max_steps': 10000, 'warmup_steps': 500, 'gradient_accumulation_steps': 1, 'max_grad_norm': 1.0, 'use_muon': True, 'num_train_timesteps': 1000, 'train_timestep_shift': 3.0, 'validation_timestep_shift': 5.0, 'snr_type': 'lognorm', 'task_type': 't2v', 'i2v_prob': 0.3, 'enable_fsdp': True, 'enable_gradient_checkpointing': True, 'sp_size': 2, 'dp_replicate': 1, 'batch_size': 2, 'num_workers': 4, 'prefetch_factor': 2, 'num_generations': 4, 'use_same_noise': False, 'output_dir': './outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting', 'checkpoints_directory': 's3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting', 'save_interval': 100, 'timer_interval': 25, 'auto_resume_from_timer': True, 'log_interval': 1, 'enable_wandb': True, 'project_name': 'SAGE-GRPO', 'run_name': 'yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting', 'wandb_key': None, 'wandb_username': None, 'wandb_base_url': 'https://api.wandb.ai', 'wandb_entity': None, 'wandb_silent': True, 'dtype': 'bf16', 'master_weight_type': 'fp32', 'seed': 42, 'validation_global_seed': 930, 'validation_interval': 0, 'validation_prompts': None, 'validate_video_length': 121, 'validation_aspect_ratio': '16:9', 'validation_fixed_size': '480x864', 'validation_num_inference_steps': 40, 'validation_negative_prompt': '', 'validation_enable_sr': False, 'validation_prompt_rewrite': False, 'validate_at_step0': False, 'resume_from_checkpoint': None, 'use_lora': False, 'lora_r': 8, 'lora_alpha': 16, 'lora_dropout': 0.0, 'lora_target_modules': None, 'pretrained_lora_path': None, 'train_video_csv': ['/nfs/ywang29/Reward_finetuning/SAGE-GRPO/assets/demo_train_decomposed_binary5.csv'], 'valid_video_csv': 'assets/demo_sample.csv', 'reward_model': 'video_score2', 'reward_config': {'models': {'video_score2': {'weight': 1.0, 'sub_reward': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}}}, 'reward_checkpoint_mode': 'v3', 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_metric': 'weighted'}, 'reward_checkpoint_mode': 'v3', 'reward_weights': {'VQ': 1.0, 'MQ': 1.0, 'TA': 1.0}, 'reward_gradient_credit': True, 'reward_credit_num_frames': 8, 'reward_credit_max_questions': 1, 'reward_credit_metric': 'weighted', 'token_credit_source_model': None, 'token_credit_advantage_source': 'token', 'token_credit_remote_url': None, 'token_credit_vlm_model_path': None, 'token_credit_vlm_model_family': None, 'token_credit_vlm_prompt_template': None, 'remote_reward_url': None, 'vlm_reward_model_path': None, 'vlm_reward_model_family': None, 'vlm_reward_prompt_template': None, 'vlm_reward_num_frames': 8, 'vlm_reward_max_pixels': 65536, 'vlm_reward_max_new_tokens': 128, 'vlm_reward_batch_size': 0, 'vlm_reward_score_type': 'logprob', 'vlm_reward_missing_score': 0.0, 'vlm_reward_video_sample_type': 'uniform', 'vlm_reward_token_credit_max_questions': 8, 'token_credit_loss_mode': 'video_adv', 'token_credit_ratio_mode': 'uniform_weighted_logprob', 'token_credit_shuffle_weights': False, 'token_credit_spatial_granularity': 'frame', 'token_credit_spatial_bins': 1, 'token_credit_spatial_window': 1, 'token_credit_chunk_size': 1, 'enable_grpo_policy_update': True, 'grpo_sampling_steps': 20, 'sample_n_frames': 81, 'eta': 0.5, 'cfg_scale': 6.0, 'neg_prompt': None, 'data_type': 'video', 'model_type': '16164', 'video_fps': 24, 'vae_name': '16164-32c-hy20250605', 'vae_tiling': False, 'infer_flow_shift_video': 5.0, 'video_bucket_hw_base_size': 480, 'video_bucket_hw_bucket_stride': 32, 'multitask_mask_training_type': 'concat', 'vision_num_semantic_tokens': 729, 'vision_states_dim': 1152, 'sde_type': 'sage_grpo', 'mini_batch_size_per_update': 1, 'clip_range': 0.0001, 'adv_clip_max': 5.0, 'timestep_fraction': 1.0, 'random_train_timestep_count': False, 'random_train_timestep_min': 1, 'random_train_timestep_max': 20, 'use_grad_balancing': True, 'enable_timestep_permutation': True, 'reference_mode_offload': True, 'kl_weight': 1e-05, 'kl_coef': 1e-07, 'kl_min_coef': 1e-07, 'use_moving_KL': True, 'update_ref_model_step': 10, 'use_dual_kl': True, 'dual_kl_moving_weight': 1.0, 'dual_kl_step_weight': 0.1, 'kl_compute_mode': 'rollout_phase', 'glyph_byT5_v2': True, 'byt5_max_length': 256, 'video_uncond_p_byt5': 0.0, 'precision': 'bf16', 'vae_precision': 'fp16', 'val_disable_autocast': False, 'debug_grad_flow': False, 'debug_train_diagnostics_interval': 1, '_wandb': {}} +2026-06-25 18:01:16,348 INFO MainThread:1812 [wandb_init.py:init():816] starting backend +2026-06-25 18:01:16,613 INFO MainThread:1812 [wandb_init.py:init():831] sending inform_init request +2026-06-25 18:01:16,944 INFO MainThread:1812 [wandb_init.py:init():836] backend started and connected +2026-06-25 18:01:16,947 INFO MainThread:1812 [wandb_init.py:init():906] updated telemetry +2026-06-25 18:01:17,001 INFO MainThread:1812 [wandb_init.py:init():929] communicating run to backend with 90.0 second timeout +2026-06-25 18:01:17,475 INFO MainThread:1812 [wandb_init.py:init():974] starting run threads in backend +2026-06-25 18:01:17,684 INFO MainThread:1812 [wandb_run.py:_console_start():2523] atexit reg +2026-06-25 18:01:17,684 INFO MainThread:1812 [wandb_run.py:_redirect():2373] redirect: wrap_raw +2026-06-25 18:01:17,684 INFO MainThread:1812 [wandb_run.py:_redirect():2442] Wrapping output streams. +2026-06-25 18:01:17,684 INFO MainThread:1812 [wandb_run.py:_redirect():2465] Redirects installed. +2026-06-25 18:01:17,686 INFO MainThread:1812 [wandb_init.py:init():1012] run started, returning control to user process diff --git a/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/run-20260529_204615-dgy7zq92/files/output.log b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/run-20260529_204615-dgy7zq92/files/output.log new file mode 100644 index 0000000000000000000000000000000000000000..ed66b4d2be00d34c6fa7d14cb8baf397bbd95845 --- /dev/null +++ b/outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/wandb/wandb/run-20260529_204615-dgy7zq92/files/output.log @@ -0,0 +1,23891 @@ +2026-05-29 20:46:16.602 | INFO | __main__:print_training_configuration:963 - SYSTEM & HARDWARE CONFIGURATION +2026-05-29 20:46:16.604 | INFO | __main__:print_training_configuration:964 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.604 | INFO | __main__:print_training_configuration:965 - Number of GPUs : 64 +2026-05-29 20:46:16.605 | INFO | __main__:print_training_configuration:966 - Local rank : 0 +2026-05-29 20:46:16.605 | INFO | __main__:print_training_configuration:967 - Global rank : 0 +2026-05-29 20:46:16.606 | INFO | __main__:print_training_configuration:968 - Device : cuda:0 +2026-05-29 20:46:16.606 | INFO | __main__:print_training_configuration:969 - Master weight dtype : torch.float32 +2026-05-29 20:46:16.606 | INFO | __main__:print_training_configuration:971 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.607 | INFO | __main__:print_training_configuration:972 - MODEL PARAMETERS & ARCHITECTURE +2026-05-29 20:46:16.607 | INFO | __main__:print_training_configuration:973 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.618 | INFO | __main__:print_training_configuration:976 - Number of total : 8,326,608,160 +2026-05-29 20:46:16.619 | INFO | __main__:print_training_configuration:977 - Total trainable parameters : 8,326,608,160 +2026-05-29 20:46:16.619 | INFO | __main__:print_training_configuration:978 - DIT model root : ./ckpts +2026-05-29 20:46:16.619 | INFO | __main__:print_training_configuration:979 - Transformer version : 480p_t2v +2026-05-29 20:46:16.620 | INFO | __main__:print_training_configuration:980 - Attention mode : flash +2026-05-29 20:46:16.620 | INFO | __main__:print_training_configuration:981 - Gradient checkpointing : True +2026-05-29 20:46:16.621 | INFO | __main__:print_training_configuration:982 - FSDP enabled : True +2026-05-29 20:46:16.621 | INFO | __main__:print_training_configuration:983 - LoRA enabled : False +2026-05-29 20:46:16.621 | INFO | __main__:print_training_configuration:984 - Glyph byT5 v2 : True +2026-05-29 20:46:16.622 | INFO | __main__:print_training_configuration:985 - ByT5 max length : 256 +2026-05-29 20:46:16.622 | INFO | __main__:print_training_configuration:987 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.623 | INFO | __main__:print_training_configuration:988 - DISTRIBUTED TRAINING CONFIGURATION +2026-05-29 20:46:16.623 | INFO | __main__:print_training_configuration:989 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.623 | INFO | __main__:print_training_configuration:990 - World size : 64 +2026-05-29 20:46:16.624 | INFO | __main__:print_training_configuration:991 - DP degree : 32 +2026-05-29 20:46:16.624 | INFO | __main__:print_training_configuration:992 - DP rank : 0 +2026-05-29 20:46:16.625 | INFO | __main__:print_training_configuration:993 - SP size : 2 +2026-05-29 20:46:16.625 | INFO | __main__:print_training_configuration:994 - SP rank : 0 +2026-05-29 20:46:16.625 | INFO | __main__:print_training_configuration:996 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.626 | INFO | __main__:print_training_configuration:997 - BATCH SIZE & DATA FLOW CONFIGURATION +2026-05-29 20:46:16.626 | INFO | __main__:print_training_configuration:998 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.627 | INFO | __main__:print_training_configuration:999 - Per GPU batch size : 2 +2026-05-29 20:46:16.627 | INFO | __main__:print_training_configuration:1000 - DP degree batch size : 64 +2026-05-29 20:46:16.627 | INFO | __main__:print_training_configuration:1001 - Gradient accumulation steps : 1 +2026-05-29 20:46:16.628 | INFO | __main__:print_training_configuration:1002 - Num generations : 4 +2026-05-29 20:46:16.628 | INFO | __main__:print_training_configuration:1003 - Use same noise : False +2026-05-29 20:46:16.629 | INFO | __main__:print_training_configuration:1004 - Num workers : 4 +2026-05-29 20:46:16.629 | INFO | __main__:print_training_configuration:1005 - Prefetch factor : 2 +2026-05-29 20:46:16.629 | INFO | __main__:print_training_configuration:1007 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.630 | INFO | __main__:print_training_configuration:1008 - OPTIMIZATION CONFIGURATION +2026-05-29 20:46:16.630 | INFO | __main__:print_training_configuration:1009 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.631 | INFO | __main__:print_training_configuration:1010 - Optimizer : muon +2026-05-29 20:46:16.631 | INFO | __main__:print_training_configuration:1011 - Learning rate : 1e-05 +2026-05-29 20:46:16.631 | INFO | __main__:print_training_configuration:1012 - Weight decay : 0.01 +2026-05-29 20:46:16.632 | INFO | __main__:print_training_configuration:1013 - Max gradient norm : 1.0 +2026-05-29 20:46:16.632 | INFO | __main__:print_training_configuration:1014 - LR scheduler : constant +2026-05-29 20:46:16.632 | INFO | __main__:print_training_configuration:1015 - Warmup steps : 500 +2026-05-29 20:46:16.633 | INFO | __main__:print_training_configuration:1016 - Max training steps : 10000 +2026-05-29 20:46:16.633 | INFO | __main__:print_training_configuration:1018 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.634 | INFO | __main__:print_training_configuration:1019 - INFERENCE & VALIDATION CONFIGURATION +2026-05-29 20:46:16.634 | INFO | __main__:print_training_configuration:1020 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1021 - Validation interval : 0 +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1022 - Validate at step 0 : False +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1023 - Validation video length : 121 +2026-05-29 20:46:16.635 | INFO | __main__:print_training_configuration:1024 - Validation global seed : 930 +2026-05-29 20:46:16.636 | INFO | __main__:print_training_configuration:1025 - Validation inference steps : 40 +2026-05-29 20:46:16.636 | INFO | __main__:print_training_configuration:1026 - Validation aspect ratio : 16:9 +2026-05-29 20:46:16.637 | INFO | __main__:print_training_configuration:1027 - Validation fixed size : 480x864 +2026-05-29 20:46:16.637 | INFO | __main__:print_training_configuration:1028 - Validation prompt rewrite : False +2026-05-29 20:46:16.638 | INFO | __main__:print_training_configuration:1029 - Validation flow shift : 5.0 +2026-05-29 20:46:16.638 | INFO | __main__:print_training_configuration:1030 - Sampling CFG scale : 6.0 +2026-05-29 20:46:16.638 | INFO | __main__:print_training_configuration:1031 - Sampling eta : 0.5 +2026-05-29 20:46:16.639 | INFO | __main__:print_training_configuration:1032 - Sampling steps (GRPO rollout) : 20 +2026-05-29 20:46:16.639 | INFO | __main__:print_training_configuration:1033 - Token credit loss mode : video_adv +2026-05-29 20:46:16.640 | INFO | __main__:print_training_configuration:1034 - Token credit ratio mode : uniform_weighted_logprob +2026-05-29 20:46:16.640 | INFO | __main__:print_training_configuration:1035 - Token credit shuffle weights : False +2026-05-29 20:46:16.640 | INFO | __main__:print_training_configuration:1036 - Token credit source model : None +2026-05-29 20:46:16.641 | INFO | __main__:print_training_configuration:1037 - Token credit advantage source : token +2026-05-29 20:46:16.641 | INFO | __main__:print_training_configuration:1038 - Random train timestep count : False +2026-05-29 20:46:16.641 | INFO | __main__:print_training_configuration:1039 - Random train timestep range : 1-20 +2026-05-29 20:46:16.642 | INFO | __main__:print_training_configuration:1043 - SDE type : sage_grpo +2026-05-29 20:46:16.642 | INFO | __main__:print_training_configuration:1045 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.643 | INFO | __main__:print_training_configuration:1046 - GRPO & KL CONFIGURATION +2026-05-29 20:46:16.643 | INFO | __main__:print_training_configuration:1047 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.643 | INFO | __main__:print_training_configuration:1048 - Reward model : video_score2 +2026-05-29 20:46:16.644 | INFO | __main__:print_training_configuration:1049 - Reward gradient credit : True +2026-05-29 20:46:16.644 | INFO | __main__:print_training_configuration:1050 - Reward credit frames : 8 +2026-05-29 20:46:16.645 | INFO | __main__:print_training_configuration:1051 - Reward credit max questions : 1 +2026-05-29 20:46:16.645 | INFO | __main__:print_training_configuration:1052 - Reward credit metric : weighted +2026-05-29 20:46:16.645 | INFO | __main__:print_training_configuration:1063 - Reward checkpoint mode : v3 +2026-05-29 20:46:16.646 | INFO | __main__:print_training_configuration:1064 - KL weight : 1e-05 +2026-05-29 20:46:16.646 | INFO | __main__:print_training_configuration:1065 - KL coef : 1e-07 +2026-05-29 20:46:16.647 | INFO | __main__:print_training_configuration:1066 - Use moving KL : True +2026-05-29 20:46:16.647 | INFO | __main__:print_training_configuration:1067 - Update ref model step : 10 +2026-05-29 20:46:16.647 | INFO | __main__:print_training_configuration:1068 - Use dual KL : True +2026-05-29 20:46:16.648 | INFO | __main__:print_training_configuration:1069 - Dual KL moving weight : 1.0 +2026-05-29 20:46:16.648 | INFO | __main__:print_training_configuration:1070 - Dual KL step weight : 0.1 +2026-05-29 20:46:16.648 | INFO | __main__:print_training_configuration:1071 - KL compute mode : rollout_phase +2026-05-29 20:46:16.649 | INFO | __main__:print_training_configuration:1072 - Use grad balancing : True +2026-05-29 20:46:16.649 | INFO | __main__:print_training_configuration:1073 - Enable timestep permutation : True +2026-05-29 20:46:16.650 | INFO | __main__:print_training_configuration:1074 - Reference mode offload : True +2026-05-29 20:46:16.650 | INFO | __main__:print_training_configuration:1076 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.651 | INFO | __main__:print_training_configuration:1077 - LOGGING & CHECKPOINTING CONFIGURATION +2026-05-29 20:46:16.651 | INFO | __main__:print_training_configuration:1078 - -------------------------------------------------------------------------------- +2026-05-29 20:46:16.651 | INFO | __main__:print_training_configuration:1079 - Output directory : ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-29 20:46:16.652 | INFO | __main__:print_training_configuration:1080 - Checkpoints directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-29 20:46:16.652 | INFO | __main__:print_training_configuration:1081 - Checkpoint interval : 100 +2026-05-29 20:46:16.653 | INFO | __main__:print_training_configuration:1082 - Timer checkpoint interval : 25 +2026-05-29 20:46:16.653 | INFO | __main__:print_training_configuration:1083 - Auto resume from timer : True +2026-05-29 20:46:16.653 | INFO | __main__:print_training_configuration:1084 - Timer checkpoint directory : s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer +2026-05-29 20:46:16.654 | INFO | __main__:print_training_configuration:1085 - Log interval : 1 +2026-05-29 20:46:16.654 | INFO | __main__:print_training_configuration:1086 - Resume checkpoint : None +2026-05-29 20:46:16.655 | INFO | __main__:print_training_configuration:1087 - Train CSV : ['/nfs/ywang29/Reward_finetuning/SAGE-GRPO/assets/demo_train_decomposed_binary5.csv'] +2026-05-29 20:46:16.655 | INFO | __main__:print_training_configuration:1088 - Valid CSV : assets/demo_sample.csv +2026-05-29 20:46:16.655 | INFO | __main__:print_training_configuration:1089 - Seed : 42 +2026-05-29 20:46:16.656 | INFO | __main__:print_training_configuration:1090 - WandB enabled : True +2026-05-29 20:46:16.656 | INFO | __main__:print_training_configuration:1092 - WandB project : SAGE-GRPO +2026-05-29 20:46:16.657 | INFO | __main__:print_training_configuration:1093 - WandB run name : yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting +2026-05-29 20:46:16.657 | INFO | __main__:__init__:951 - [Effective Config] world_size=64, rank=0, local_rank=0, sp_size=2, dp_size=32, dp_rank=0, sp_rank=0, batch_size=2, num_generations=4, grad_accum=1 +2026-05-29 20:46:16.657 | INFO | __main__:_build_video_dataset:2038 - [Effective GRPO Grouping] video_batch_size=2, dp_replicas=32, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, product=64 +2026-05-29 20:46:17.315 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1178 - [Config Validation] video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1 +2026-05-29 20:46:17.318 | INFO | hyvideo.pipelines.hunyuan_video_grpo_pipeline:get_post_train_video_dataloader:1205 - Dataloader init done +2026-05-29 20:46:17.319 | INFO | __main__:_train_grpo:2121 - Starting GRPO training... +2026-05-29 20:46:17.319 | INFO | __main__:_train_grpo:2122 - Max steps: 10000 +2026-05-29 20:46:17.320 | INFO | __main__:_train_grpo:2123 - Batch size: 2 +2026-05-29 20:46:17.320 | INFO | __main__:_train_grpo:2124 - Learning rate: 1e-05 +2026-05-29 20:46:17.512 | INFO | __main__:_resolve_resume_checkpoint:1820 - No timer checkpoint found at s3://snap-genvid-us-east-2/ywang29/checkpoints_large_model/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/timer; starting from scratch. +Iterator of DistributedSamplerWithStartIndex created. +2026-05-29 20:46:19.144 | INFO | __main__:_train_grpo:2151 - [FLAG] post_train train path -> train_one_step +2026-05-29 20:51:46.102 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1036 - [ParallelGroups] Rank 0, dp_rank 0: video_batch_size=2, num_generations=4, ranks_per_group=2, num_groups_per_rank=1, group_idx=0, rank_in_group=0 +2026-05-29 20:51:46.216 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 1/2 (samples 1-1/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:51:46.220 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +20it [01:17, 3.88s/it][01:17<00:00, 2.99s/it] +100%|██████████| 20/20 [01:17<00:00, 3.88s/it] +2026-05-29 20:53:12.758 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28576_dp_0_rank_0_sample_0_slot_0.mp4 +qwen-vl-utils using torchcodec to read video. +video_reader_backend torchcodec error, use torchvision as default, msg: Could not load libtorchcodec. Likely causes: + 1. FFmpeg is not properly installed in your environment. We support + versions 4, 5, 6, 7, and 8, and we attempt to load libtorchcodec + for each of those versions. Errors for versions not installed on + your system are expected; only the error for your installed FFmpeg + version is relevant. On Windows, ensure you've installed the + "full-shared" version which ships DLLs. + 2. The PyTorch version (2.10.0+cu130) is not compatible with + this version of TorchCodec. Refer to the version compatibility + table: + https://github.com/pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec. + 3. Another runtime dependency; see exceptions below. + + The following exceptions were raised as we tried to load libtorchcodec: + +[start of libtorchcodec loading traceback] +FFmpeg version 8: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.60: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core8.so + +FFmpeg version 7: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.59: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core7.so + +FFmpeg version 6: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.58: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core6.so + +FFmpeg version 5: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.57: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core5.so + +FFmpeg version 4: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core4.so: undefined symbol: torch_from_blob + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core4.so +[end of libtorchcodec loading traceback]. +/usr/local/lib/python3.12/site-packages/torchvision/io/_video_deprecation_warning.py:9: UserWarning: The video decoding and encoding capabilities of torchvision are deprecated from version 0.22 and will be removed in version 0.24. We recommend that you migrate to TorchCodec, where we'll consolidate the future decoding/encoding capabilities of PyTorch: https://github.com/pytorch/torchcodec + warnings.warn( +The following generation flags are not valid and may be ignored: ['temperature']. Set `TRANSFORMERS_VERBOSITY=info` for more details. +2026-05-29 20:53:30.323 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1173 - Rank 0 generating batch 2/2 (samples 2-2/2), eta: 0.5, batch_size: 1, 81x352x624, flow_shift: 5.0, infer_steps: 20 +2026-05-29 20:53:30.326 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1198 - Rank 0 using 'all' strategy (no mixgrpo): determistic=False for all timesteps +20it [01:01, 3.05s/it][01:01<00:00, 2.95s/it] +100%|██████████| 20/20 [01:01<00:00, 3.05s/it] +2026-05-29 20:54:38.379 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1331 - Rank 0: Saved video 1/1 to ./outputs/yw008-2-sage-videoscore2-token-credit-video-adv-no-weighting/rl_samples/0000000/group_0/index_7542_seed_28577_dp_0_rank_0_sample_1_slot_1.mp4 +video_reader_backend torchcodec error, use torchvision as default, msg: Could not load libtorchcodec. Likely causes: + 1. FFmpeg is not properly installed in your environment. We support + versions 4, 5, 6, 7, and 8, and we attempt to load libtorchcodec + for each of those versions. Errors for versions not installed on + your system are expected; only the error for your installed FFmpeg + version is relevant. On Windows, ensure you've installed the + "full-shared" version which ships DLLs. + 2. The PyTorch version (2.10.0+cu130) is not compatible with + this version of TorchCodec. Refer to the version compatibility + table: + https://github.com/pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec. + 3. Another runtime dependency; see exceptions below. + + The following exceptions were raised as we tried to load libtorchcodec: + +[start of libtorchcodec loading traceback] +FFmpeg version 8: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.60: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core8.so + +FFmpeg version 7: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.59: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core7.so + +FFmpeg version 6: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.58: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core6.so + +FFmpeg version 5: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: libavutil.so.57: cannot open shared object file: No such file or directory + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core5.so + +FFmpeg version 4: +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1442, in load_library + ctypes.CDLL(path) + File "/usr/local/lib/python3.12/ctypes/__init__.py", line 379, in __init__ + self._handle = _dlopen(self._name, mode) + ^^^^^^^^^^^^^^^^^^^^^^^^^ +OSError: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core4.so: undefined symbol: torch_from_blob + +The above exception was the direct cause of the following exception: + +Traceback (most recent call last): + File "/usr/local/lib/python3.12/site-packages/torchcodec/_internally_replaced_utils.py", line 93, in load_torchcodec_shared_libraries + torch.ops.load_library(core_library_path) + File "/usr/local/lib/python3.12/site-packages/torch/_ops.py", line 1444, in load_library + raise OSError(f"Could not load this library: {path}") from e +OSError: Could not load this library: /usr/local/lib/python3.12/site-packages/torchcodec/libtorchcodec_core4.so +[end of libtorchcodec loading traceback]. +The following generation flags are not valid and may be ignored: ['temperature']. Set `TRANSFORMERS_VERBOSITY=info` for more details. +2026-05-29 20:54:59.835 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:808 - [GatherRewards] Rank 0: Using local metric names without global sync: ['avg', 'video_score2_mq', 'video_score2_overall', 'video_score2_ta', 'video_score2_vq'] +/usr/local/lib/python3.12/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning. + return func(*args, **kwargs) +2026-05-29 20:55:07.137 | INFO | hyvideo.utils.grpo_utils:gather_and_process_rewards:876 - [GatherRewards] Rank 0 saved 4 entries for group 0 +2026-05-29 20:55:17.523 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1566 - Rank 0: Pre-computing Reference Model statistics (Batch-wise)... +Processing reference model mini-batches: 100%|██████████| 2/2 [01:35<00:00, 47.91s/it] +2026-05-29 20:57:22.660 | INFO | hyvideo.utils.grpo_utils:prepare_samples_online:1709 - Rank 0: Reference stats computed (Batched). Shape: torch.Size([2, 20, 32, 21, 22, 39]), Time: 124.72s (2.08min) +2026-05-29 20:57:23.798 | INFO | hyvideo.utils.grpo_utils:train_one_step:2405 - Rank 0, sp_rank 0: token_credit GRPO active with 2 valid question slots; loss_mode=video_adv, ratio_mode=uniform_weighted_logprob, advantage_source=token, shuffle_weights=False, uniform_weights=True. +2026-05-29 20:59:23.338 | INFO | hyvideo.utils.grpo_utils:train_one_step:2620 - [All Timesteps] Training on timesteps: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19] +Global Step 0: training: 0%| | 0/2 [00:00